CA2049758C - Optical character recognition system and method - Google Patents
Optical character recognition system and methodInfo
- Publication number
- CA2049758C CA2049758C CA002049758A CA2049758A CA2049758C CA 2049758 C CA2049758 C CA 2049758C CA 002049758 A CA002049758 A CA 002049758A CA 2049758 A CA2049758 A CA 2049758A CA 2049758 C CA2049758 C CA 2049758C
- Authority
- CA
- Canada
- Prior art keywords
- character
- mismatch
- predetermined
- pattern
- producing
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Expired - Fee Related
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/74—Image or video pattern matching; Proximity measures in feature spaces
- G06V10/75—Organisation of the matching processes, e.g. simultaneous or sequential comparisons of image or video features; Coarse-fine approaches, e.g. multi-scale approaches; using context analysis; Selection of dictionaries
- G06V10/751—Comparing pixel values or logical combinations thereof, or feature values having positional relevance, e.g. template matching
Abstract
A system and method are disclosed for optical scanning and recognition of alphanumeric characters of different sizes and fonts such as E13B, OCRA, OCRB, and Farington 7B type fonts recorded on documents such as bank drafts and checks. The system comprises an optical scanner, a microprocessor, a read-only-memory (ROM), a random-access-memory (RAM, with a stored program) for storing data and a plurality of predetermined character-identification patterns (templates). Each template includes three configurations: an actual pixel pattern (bit configuration) of a character, a configuration representing significant portions of the character which remain the same and do not change despite changes in size of the character, and a configuration representing portions of the character which are given added weight in distinguishing between similar characters. The scanner optically scans each document and produces a plurality of gray-scale pixel values which are stored in the RAM under control of the microprocessor. The microprocessor processes the stored pixel values, and effectively locates and segments each character on the document. The microprocessor then converts the segmented-character pixel values to binary data corresponding to the intervals, and selects, from the plurality of templates, the template that matches the binary data, the matching template serving to identify (recognize) the segmented character.
Description
: ~ p~ ~, ~ OPTICAL CHARACTER RECOGNITION SYSTEM AND METHOD
:
Background of the Invention This invention relates to the scanning of documents and the recognition of alphanumeric symbols generally, and particularly to optical scanning and recognition of alphanumeric characters such as E13B, OCRA, OCRB, and Farington 7B type fonts.
The E13B, OCRA, OCRB, and Farington 7B
character sets have received widespread use in financial transactions in the United States and abroad. Generally, the characters are recorded on documents such as bank drafts, checks, and other financial documents. For example, E13B type fonts are generally recorded in MICR ~magnetic ink character record) form for scanning by a magnetic scanner.
A vidicon camera system is sometimes used to scan the document and to record an image of it for archival and evidentiary purposes while, in a separate ~ operation, a magnetic scanner system is used to read ; the magnetic-ink characters (e.g., E13B characters).
This need for two separate scanner systems to perform two separate operations presents a costly, time-consuming problem. Further, hardware for processing E13B, OCRA, OCRB and Farrington 7B type character is often complex and expensive, and unduly time-consuming in performing cbaracter recognition. Also, such hardware often binarizes the entire image, introducing noise into the binarized image and causing loss of image data, making the recognition process more dif f icult.
A way of providing for image scanning and character recognition that would eliminate such a two-system two-operation requirement, and that would significantly reduce noise and image-processing time, is needed and would be useful.
~'- .' '~ ' :~ .
. .. - .. .. .. , ...... . . ., . .- ,.. .- . j ~.. - , , -, , ,~ . .
,, ~ f~
Summary of the Invention Accordingly, a system and method for optical character recognition are provided which obviate the need for separate magnetic scanner systems and the performance of separate magnetic scanning operations, and reduces noise and image-processing time without causing loss of image data.
In a first aspect of the invention, a system (apparatus) is provided comprising: (a) a CCD scanner for optically scanning an alphanumeric character on a document, the character and document having different opacities, and producing a plurality of gray-scale pixel values corresponding to said opacities, (b) a memory disposed for storing the gray-scale pixel values, a program, and a plurality of predetermined binary patterns, one of said patterns representing the character, and (c) a processor coupled to the scanner ~.
and to the memory, and operating under control of the program, for locating the character on the document by selecting from said gray-scale pixel values a matrix of values representative of the character, for converting the matrix values to binary data, and for selecting from the plurality of predetermined binary patterns a pattern that matches the converted binary data, said matching pattern serving to identify the character.
The CCD scanner includes a control logic, the processor includes a microprocessor, and the memory includes a ROM (read only memory) with a boot routine and a RAM (random access memory) with the program, the pixel values, and the binary patterns.
In a second aspect of the invention, a method is provided for optically identifying an alphanumeric character on a document, the character ~.
and document having different opacities, the character -;.
being represented by one of a plurality of :
.
,, _ .
~ :; 8 predetermined binary patterns, the method comprising the steps of: (a) optically scanning the character on the document and producing a plurality of gray-scale pixel values corresponding to said opacities, (b) locating the character by selecting, from said gray-scale pixel values, a matrix of values representative of the character, (c) converting the matrix values to binary data, and (d) selecting from the plurality of predetermined binary patterns a pattern that matches the converted binary data, said matching pattern identifying the character.
The characters are recorded on the document in, for example, E13B format. The camera (CCD
scanner) scans the document vertically, from bottom to top, starting at the bottom right corner of the document. The document is moved to the right continuously (with the width of each vertical scan corresponding to the diameter of one pixel) until the top left corner of the document is reached indicating that scanning has been completed.
In locating a character (i.e., the gray-scale pixel values representing the character) for proce~sing, the system positions a window over selected pixel values of the character (thereby framing the character) such that the sum of the selected pixel values within the window is a maximum.
The system then determines the arithmetic mean of the dark pixels within the window. It does this by determining the arithmetic mean of pre-selected ones of the selected pixel values (the preselected ones representing pixel values above a predetermined threshold, the predetermined threshold representing a value greater than the character's background level (determined by pre-scan of the character). The system then converts the selected pixel values to binary data relative to said mean (i.e., converts to one or zero depending on whether the magnitude of each selected :
pixel value is less than, or is greater than or equal to, said mean). The binary data is then logically compared to each of the predetermined binary patterns until a matching pattern is found which best identifies the character. The predetermined binary patterns represent binary templates, each pattern (template) having three bit configurations (levels), a first bit configuration representing the actual bit pattern of the character, a second bit configuration representing the significance of the bits of the actual bit pattern, and a third bit configuratlon representing weighted values for the bits of the actual bit pattern. The extent to which the converted binary data logically compares to each of the three bit configurations of each of the predetermined bit patterns is expressed as mismatch words, from which mismatch counts are calculated. The mismatch counts ;-are then compared to at least one predetermined threshold value, and the lowest mismatch count that is below (less than) the threshold value i8 selected as representative of the character.
Thus, as indicated above, the system performs segmentation of characters from the gray-scale image and binarizes the segmented characters for subsequent recognition, rather than binarizing the entire image for recognition of characters as do prior-art systems. Also, whereas such prior binarization of the entire image introduced undue amounts of error (noise) into the scanned data, the method of the present invention reduces such errors by using gray-scale values in the character-segmentation process. Further, the system is able to recognize different fonts ~i.e., different characters at different resolutions) merely by storing template tables for that font. -. ''-'' .
: .:
..
:~ .
... . - . --~: ~. . . . . .
. . - .. . . ., . , .. ,, . .: i ,.. . ... . . .. . .
- 5 - ~ ,t'l~.
Brief Description of the Drawings Figure 1 is a block diagram of the system of the present invention;
Figure 2 is a graphic illustration of the appearance of a character scanable and identifiable by the system of Figure l;
Figure 3 is a graphic illustration of a portion of a document with various characters recorded thereon;
Figure 4 is a graphic illustration of a scanned character having a matrix of pixels of different opacities (gray-scale values);
Figure 5 is a graphic illustration of a window used for framing a character;
Figure 6 iB a graphic illustration of the window of Figure 5 used in framing the character of Figure 4;
Figure 7 is a block diagram showing the results of various logical operations performed on binary data representing the scanned character;
Figure 8 is a block diagram of a matrix repre~enting various portions of the image of a framed character; and Figure 9 is a graphic/tabular illustration .-of a template and associated binary values used by the system of Figure 1.
Description of the Preferred Embodiment Referring now to Fig. 1, there is shown a system (apparatus) 11 of the present invention. The system includes and optical reader or scanner 13 for : :
scanning a document 19, and a computer system 21 for proce~sing data received from the scanner 13. The sC~nner may be a camera with a CCD (charge coupled device) 15 and a controller 17. The document 19 may be a bank draft or check with numeric character or .
.; .
other alphanumeric symbol 20 recorded thereon. The computer system 21 comprises a microprocessor 23, a ROM (read-only-memory) 25 with a boot program, and a RAM (random access memory) 27, with a stored program, for storing data from the scanning device and from the microprocessor and for storing predefined data.
The scanning operation is commenced at the bottom right corner 33 of the document l9 and ends at the top left corner 35. The scanner 13 scans the document l9 vertically, from bottom-to-top, as shown by arrow 37, first along right-most virtual scan line 39, then again from bottom-to-top along next-right-most virtual scan line 41, and so on until the top left corner of the document is reached. The virtual scan lines are arranged in parallel. The document 19 is moved continuously from left to right, as shown by arrow 43, while the scanner scans along successive scan lines, the width of each scan line corresponding to the diameter of a pixel (approximately .0127 cm or .005 inch).
An example of a read line of a document i9 shown in Figure 3. When a document is scanned, the pixel values generated by the scanner to represent the document, including the characters thereon, are stored in memory 27 (Figure 1). The characters are each stored as a data array (matrix) having, for example, typically one hundred rows, corresponding to the read line's height, and twelve hundred columns corresponding to the document'~ length (twelve hundred being the typical number of columns for a six-inch document at 200 pixels per inch). Within the data array, a character (when one is located and segmented as described below) is represented by a matrix of rows (e.g., twenty-one, corresponding to the character's height) and columns (e.g., fourteen, corresponding to the character's width). The character "three" shown in Figure 2, for example, may be represented in pixel . , .
. ~ .
array form as depicted in Figure 4. Each pixel 49 represents a portion of the document and has an opacity index or gray-scale value in the ran~e 0-255.
The dark pixels generally have a greater gray-scale value (approaching 255), and the light pixels generally have a lesser gray-scale value (approaching 0). Since characters are normally printed in a dark color compared to the generally light colored background of the document, the difference in opacity between the character and its background is easily determinable.
As indicated above, these gray-scale data values are generated by the scanner, upon scanning the document. The generated data are then stored in memory 27 (Figure 1) for processing by the system 21.
The system, under control of the stored program, examines the stored data values, and proceeds to locate a first character, then successive characters of a first field, then other characters of successive fields. Upon locating a character, the system proceeds to frame or delimit the character (i.e., to separate-out or segment the character from other characters). Starting with the stored gray-scale data values obtained from scanning the document from bottom-to-top along the right-most scan line 39 (Figures 1 and 2), and then from bottom-to-top along the next-right-most scan line 41, and so on, the system locates and frames (segments) each of the characters as described below. It does this by first searching the stored data for a rise in gray-scale value representing a transition from a light pixel to a dark pixel, vertically from bottom to top along the scan line. If a rise of more than, for example, forty the particular value, such as forty, being font dependent) is found between two gray-scale values of adjacent pixels in the scan direction 37, a dark pixel is deemed to be found. The region (of the document) proximate to the dark pixel is regarded as a region where a character is likely to be found. When a dark pixel is found, the system checks the character region (i.e., checks the stored data proximate to the dark pixel value) to determine if the dark pixel is part of a character. It checks for the presence of a character by establishing a segmentation window of sufficient size to effectively frame the character on the document, the position of the dark pixel corresponding to the center of the window. As depicted in Figure S, a segmentation window 53 is shown having left and right edges (columns) 55,57, and top and bottom edges (rows) 59,61. Bordering the left and right edges are columns (left+l) 63 and (right-l) 65, and bordering the top and bottom edges are rows (top+l) 67 and (bottom-l) 69.
As explained below with reference to Figures S and 6, a character is considered framed and positioned in the center of the window when the sum (Sw) of all the gray-scale pixel values in the window is maximum. To determine whether the sum (Sw) increases when the window is moved, the sums of the pixel values in each of the left and right columns 55,57 and in each of their bordering columns 63,65 are calculated. Likewise, the sums of the pixel values in each of the top and bottom columns 59,61 and in each of their bordering columns 67,69 are calculated. If, -for example, the sum in column 63 is found to be greater than the sum in column 57, and would cause the sum (Sw) in the window to increase when the window 53 is moved by one pixel leftward to include column 63 and exclude column 57, then the window is moved leftward. Likewise, if by moving the window by one pixel upward to include row 67 and exclude row 61, the sum (Sw) in the window would increase, then the window is moved upward. Thus, from the sum (cumulative count) of the pixel values of the respective rows and .
' h, ' . , . ~ . . . , ,. ' . .. ' . . ,, '' .. ' . , . ' . ' ', ~ , : : ' ' . :
columns adjoining (bordering) the window, relative to the sum of the pixel values of the respective rows and columns at the edges of the window, the direction can be determined in which the window can be moved to increase the sum (sw) of the pixel values within the window. As shown in Figure 6, according to the following counts:
Right-l column count = 0 Right column count = 0 Left column count = 100 Left+l column count = 100 Bottom-l row count = 0 Bottom row count = 0 Top row count = 170 Top+l row count = 170 the window count is increased by 100 upon a move leftward, and increased by 170 upon a move upward.
Therefore, the window is moved leftward and upward.
Thus, the window is moved in the direction which increases Sw, and this operation is repeated until no movement of the window in any of the four directions (left, right, up, down) increases Sw. The window is then considered to have framed (delimited/segmented) a character, with the character positioned at the center of the window. Thus, by means of the window a matrix of pixel values most representative of the character is selected (framed).
As a check to determine if there is enough information (dark-pixel gray-scale values) in the window to represent the framed character, the total number of pixels in the window that exceeds a predetermined confirmation threshold (e.g., a threshold gray-scale value of 100) is obtained. If the total number i~ greater than, for example, sixty (for the E13B font), then a character is confirmed to be found in the window. Otherwise, the object framed in the window is interpreted as noise, and the system proceeds to find the next character.
,, .
, - ::, ~ . , . . ,, . , , , . . . " . . , , .. "- . . - , ~ .
~ , , , . ~ . . , . . - . . . . .. " , , ., . , ... . - - . .
., .. .~, ,., . . . : .. .. . .. ..
,f ~
: . , After segmentation, the segmented (framed) character is scaled in preparation for identification (recognition). Sixteen bit words are used in the recognition process (i.e., width of template is 16 pixels). Because many characters are wider than 16 pixels (including E13B characters which are 19 pixels wide), they are converted to 16 bits after segmentation. This is accomplished by eliminating predetermined columns e.g., for the E13B font, columns 4, 9 & 14 are eliminated. (OCRA and OCRB fonts are 16 pixels wide, at two hundred pixels per inch, and do not need to have columns eliminated.) ~ The preceding describes how a character is ; located in preparation for processing, i.e., how it is effectively located on a document and framed (segmented) for later processing. The following describes how the framed character is converted to binary form then matched to a template (predetermined binary pattern) for the purpose of identifying (classifying/recognizing) the character.
In the conversion operation, the stored ;
gray-scale pixel values for each row of the framed character are converted to binary form. First a reference threshold value i~ calculated as the ~;
arithmetic mean of the dark pixels of the framed character (as previously described). Then, the gray- ~-scale pixel values are converted to binary form relative to the reference threshold. For example, if the calculated reference threshold is one hundred, then, using the binary equivalence scale ~hown below:
. . .
Gray-Scale Pixel Values Binary Equivalent !'' > 100 1 ( black) < 100 0 (white) a row of gray scale values of 80, 120, 130, 90, 85, 70, 110, 135 may be converted to binary data having the bit pattern 01100011, as shown in the eight-bit word 71 o~ Figure 7. , ' This binary data (which represents a row or slice of the framed character) and other rows of converted binary data representing the character are then compared to a plurality of templates, each template corresponding to a character. Each template represents a predetermined pattern of binary data and comprises three distinct layers (bit configurations).
The three eight-bit words 73, 75, 77 shown in Figure 7 represent an eight-bit row in each of the three layers. The first layer (bit-pattern layer, corresponding to pattern word 73) represents the actual black/white (0/1) bit pattern of the character.
The second layer (bit-significance layer, corresponding to bit-significance word 75) identifies the bit positions of the character that are significant and those that are not significant. One-bits identify significant bit positions where the bit pattern of the character is expected to remain unchanged regardless of small changes in size and shape of the character. Zero-bits identify non-significant bit positions where the bit pattern need not be the same for different sizes or shapes of the character. The third layer (bit-weight layer, corresponding to weight word 77) identifies which bits are critical, and hence have greater weight than others, in distinguishing between like (substantially similar) characters. In this third layer, l-bits are designated as weighted bits, and 0-bits are designated as non-weighted bits. For example, l-bits in the top row of the third layer of the template for the characters "O" and "U" would be useful in distinguishing between the two characters, and l-bits in the bottom row of the third layer of the templates for the characters "Q" and "O" would be useful in distinguishing between the Q and the O.
; As shown in Figure 7, various logic operations are performed by processor 23 (Figure 1) on - . . ... . .
; ~
the binary data 71 in order to identify the character.
(The logic operations are actually performed as sixteen-bit operations. However, for simplicicity, eight-bit operations are shown). First, the binary data is EXCLUSIVE-OR'd with the corresponding word 73 of the bit-pattern layer to produce a pattern of mismatch 74 between the binary data 71 and the predetermined correct pattern 73. This mismatch pattern 74 is then logically ANDed with the corresponding word 75 of the bit-significance layer for producing a mismatch word 76, representing the extent of mismatch independent of character size and shape. The mismatch word 76 is then weighted, i.e., logically ANDed with the corresponding word 77 in the bit-weight layer to produce a weighted mismatched indicator 78. (If weighting word 77 contained no 1-bits, no weighting or ANDing would have been performed (in order to save processing time since the outcome would be zero), and the system would have proceeded -with processing the next row of binary data.) Following the weighting operation, a mismatch count (MCR) is calculated for the row of binary data 71. This calculation is performed by processor ~3 (Figure 1) as follows:
MCR = MCw + (WOC x WF) (1) where, MCW represents a mismatch count for the mismatchword (i.e., the number of ones -:
found in mismatch word 76), WOC represents a mismatch indicator or weight out-put count (i.e., the number of ones found in mismatch indicator'!' ' word 78), and WF represents a predetermined weighing factor (e.g., the integer 2 for E13B
type characters).
.:
., :
.~ .
Thus, for the mismatch bit pattern shown in the words 76 and 78 in Figure 7, it can be seen that MCW = 2, WOC = 1, and WF = 2. Therefore, the mismatch count (MCR) calculated for the row of binary data 71 would equal four (i.e., 2+(1x2) according to equation 1 above).
After calculating the mismatch count for one row, the system proceeds to calculate, as indicated above, the mismatch count for all remaining rows of the character. The mismatch counts for all the rows of the character are then summed to produce a template mismatch count (i.e., a mismatch count for the template as applied to the framed character). In a similar manner, for the other templates of the character set for the font being processed, a mismatch count is produced for each template relative to the framed character. The two lowest template mismatch counts and their associated template numbers are stored in memory as the various templates are processed. The criterion for identifying a character is: if the template with the lowest count is below a threshold (a reje~t threshold of 40 for E13B fonts) and is lower than the next lowest template by a predetermined amount ~a difference threshold of 5 for El3~ fonts), it identifies the character. ~ -Template recognition, as described previously, can be performed in a total of nine positions as shown in Figure 8. This is done in order to identify characters that may not be perfectly framed due to factors like noise in the image. In the example shown in Figure 8, the template and input (binary data) pattern sizes are 23 rows (high) by 16 columns (wide). The three layers of the template (pattern layer, bit significance layer and weighting layer) are all shifted (moved) the same amount and in the same direction. In the description of shift positions presented below, template row refer~ to the ~-'~ c~
bit pattern row, bit significance row and bit weight row.
In the center position 87, the template 96 is placed directly over the input pattern 98. Only rows 2 through 22 are used in this match. Thus, template row l is compared to pattern row 1, template row two is compared to pattern row 2 and so on. All columns (l through 16) are used in the center horizontal positions 81, 87, 93. In the upper center position 81, rows l through 22 of the template are compared to rows 2 through 23 of the input pattern.
This is the same as shifting the template pattern vertically upwards by one row. In this case, template row l is compared to input pattern row 2, template row 2 i9 compared to input pattern row 3 etc. All columns of the template and input pattern are compared in this situation (the horizontal center). Similarly, in the lower center position 93, the template is shifted down by one row, thus matching template rows 2 through 23 against input pattern rows l through 22.
The horizontal shift positions correspond to :
a shifting of the template 96 to the left or right over the input pattern 98. In the center left position 85, template columns l through 15 are used ;-and input pattern columns 2 through 16 are used.
(Since it is a center position vertically, rows 2 through 22 are used for both the input pattern and the template.) Thus, template word bit l is compared to pattern word bit 2, template word bit 3 is compared to pattern word bit 4, and so on. ~ -. . .
For example, if the template pattern layer word is: ' 0011111100001111 .: ' and the input character pattern word is:
0010111100011110 ~:'.'' The template is shifted left l bit (column):
0011111100001111 ~:
' .
.~, . .
' ,":-and the result of the exclusive Or operation would be:
It should be noted that when a shift left operation is performed, the rightmost bit (least significant bit, LSB) is 0. Thus, the LSB of the bit-significant word will be 0 also (since all three layers are shifted in the same direction and amount) and, as such, the LSB
of the mismatch word 76 (after the AND operation with the bit-significant word) will always be 0.
Similarly, a shift to the right makes the leftmost bit (most significant bit, MSB) a 0 and, as such, the MSB
of the mismatch word 76 (after the AND operation) will always be 0.
A horizontal shift to the right (position 89) is similar to the left shift, only in the opposite direction. Thus, bit 2 of the template word would line up (be compared to) bit 1 of the input pattern word.
The corner positions (positions 95, 91, 83, 79) represent a combination of a vertical shift of one row and a horziontal shift of one column. A~ an example, in the upper left position 79, the template row 1 is shifted left by one bit position and compared to pattern word 2, and template row 2 is shifted left by one bit position and compared to pattern word 3, and so on.
To recognize a character using this method requires nine passes per template times the number of templates in the character set. For example, an alphanumeric character set with 49 characters would require a total of 441 (9*49) matches being done.
Thi9 can be too time consuming. To promote faster character identification, only the center position of each template is compared initially to the input pattern. The lowest two mismatch counts and their corresponding template numbers are stored in memory 27 ~-.
.. .. .
as the various templates are processed. At the end of this first pass, the other eight positions of only these two templates are processed to find the best possible (i.e., lowest) total mismatch against the input pattern. Thu~, only 6S (1*49 + 2*8) matches need to be made ordinarily (i.e., about 15% of the 441 matches that were required).
A description of the criteria for accepting or rejecting a character i5 presented below.
The lowest template mismatch count is compared to a predetermined reject threshold for the character that the template represents. If the template mismatch count for this character is below this threshold and the difference in template mismatch counts for the lowest two templates (second lowest mismatch minus lowest mi~match) is greater than a predetermined amount (e.g., 10), the template with the lowest template mismatch identifies the framed ;
character. The reason for checking if the mismatch counts are too close is done to discriminate between characters that are very much alike (e.g., Q and 0).
If the mismatches are similar, the character is rejected rather than risking a misidentification (misread). If no template mismatch falls below the reject threshold discussed above or the two template mismatch counts are too close together, a post processing operation (described below) is performed.
If after the post-processing operation the framed character is still rejected, its image is enhanced and converted to binary form (as described below). The recognition process (described above) i~ then performed again. If the character i~ identified (satisfies the above criteria), the system then proceeds to process (locate and frame) the next character as described above.
By examining the center positions only, the pos~ibility may be missed that a character, other than ,:, ., , , , . ,; .. . .. . . . . . . . . .
the closest two, may have a closer count in a position other than the center and be the correct choice. The post processing operation can remedy such a situation.
If the lowest two characters chosen are both over the predetermined reject threshold for that individual character or the inter-character distance (second lowest mismatch-lowest mismatch) is too small, a number of other templates may be tested in the other eight positions in an attempt to find a better match.
The choice of which templates to process is made, for example, based on the closest character's similarity to the templates. For example, the closest two characters (those with the lowest mis~atch counts) chosen, the letters D and O, both have mismatch counts above their individual reject threshold. In post processing operation, a predetermined set of characters that are similar to D and O would then be processed. This could include characters 0 (zero~, Q, C and U. The mismatch counts of the closest two characters after this operation are checked (the one with the lowest mismatch count is checked first) to see if they meet the reject criterion and inter-character distance criterion. If at least one character meets the criteria, that character is returned as the result, and processing (locating and framing) of the next character in the image proceeds.
If neither of the above characters meets the reject criteria, or the inter-character distance is too small, the framed character is rejected. In this case, its image is enhanced, (as described below), the character is rebinarized, and the recognition process described above is repeated a second time. If ~he character is recognized, it is returned. Otherwis~, a reject character code is returned as the result, and processing (locating and framing) of the next character on the document continùes.
., .
In addition to the reject threshold described above, an exit threshold (e.g., the value 10) may also be used in the character identification process as described below. To enhance processing speed, successive template mismatch counts are compared to the exit threshold value, and the first template with a template mismatch count that fall~ :
below the exit threshold value would identify the character. In which event, the remaining templates are not processed (which enhances processing speed).
Further, to speed the character identification process, if, while the individual rows of a template are being processed, the mismatch count of the current template (i.e., the template currently being processed) is found to be larger than the second-lowest-count-obtained-thus-far, processing of the current template may be discontinued since this current template is no longer a contender to be one of the two with the lowest template mismatch counts. To further enhance recognition speed, the templates may be presequenced in the order in which they are most likely to occur (e.~., numbers before vowels followed by consonants in an alphanumeric font). If the mismatch is below the exit threshold (as mentioned above), the remainder of the templates need not be examined. Thus those characters occurring most frequently should be examined first. Also, to promote faster character identification, only the center position may be tested by the templates, after which, if the lowest template mismatch count is above the exit threshold, the remaining eight positions for each of the two template~ closest to the exit threshold may be processed, and the template with the lowest template mismatch count selected a~ representative of the character. Further, to speed identification, the remaining eight positions (Figure 8) may be processed in the following sequence 79, 81, 83, 85, 89, 91, 93, 95. This is the order in which the character is most likely to be positioned.
After identifying the framed character, the system proceeds to process the next character (i.e., to process the next group of stored pixel values representing the region on the document proximate to the framed character) by moving the segmentation window a selected number of pixels (e.g., three pixels) to the left of the previously framed character. The character in this proximate region is then framed (segmented) and identified as described above. All of the remaining characters (remaining stored pixel data) are processed in this manner until end-of-field is detected (i.e., no more characters have been located or an end-of-field character has been located).
If, during the character-identification process a character i9 rejected, its image is enhanced to facilitate identification. The image may be enhanced, for example, by producing a character image having constant stroke widths (i.e., having a constant number of dark pixels across the width of each vertical stroke or horizontal stroke of the character). This is accomplished as described below.
First, the gray-scale pix~l values (e.g., 6-bit, 8-bit, or 16-bit values) representing the character are converted to a lesser range of values (e.g., 3-bit gray-scale pixel values for faster processing), then the individual pixels of the character image are examined and only those pixels that are the darkest two or three pixels of the stroke are turned black, others are turned white. This produced a binary image with constant stroke widths as described for example in U.S. Patent No. 4,625,330, incorporated by reference herein).
As an assist for improving character identification, the edges of each character may be : ' .
processed as described below. As mentioned above, each template used in the character identification process corresponds to one character of the selected -character set (e.g., the E13B character set) and, as shown in Figures 2 and 4, each character is fourteen pixels wide by twenty-one pixels high. One extra column is added to the left and to the right of the character, and one extra row is added to the top and to the bottom of the character. This results in a character size of 16 by 23. Corresponding to each character, each template is sixteen pixels wide by twenty-three pixels high. Each template has an extra column on each side (right side and left side) o~ the fourteen-column character, and one extra row on the top and an extra row on the bottom of the twenty-one-row character. Each extra row comprises sixteen white pixels, and each extra column comprises twenty-three white pixels. The extra rows and columns serve to surround the 14 x 21 character region with a border of white pixels for better identification of the edge portions of certain characters, as explained below.
For example, if the character "E" is matched against a 14 x 21 "F" template, and the template is shifted upward one pixel position, the pixel values at the bottom of the "E" would be lost since it could not be used in generating the template mismatch count. This could result in mis-identification of the E as an F
since the mismatch count would be low. With the 16 x 23 template, the bottom of the E would not be lost and would contribute substantially to the mismatch count against (relative to) the bottom white row of the F
template, thereby avoiding mis-identification of the E
as an F. A similar explanation holds with respect to the right and left column positions of the 16 x 23 template, the columns serve to reduce mis-identification between characters such as "B" and "3".
., .
:
.
. . . .. . . . ~ -- . . - ..
As shown in Figure 9, each 16 x 23 template may be represented by sixty-nine sixteen-bit words of RAM 27 (corresponding to a scan of two hundred pixels per inch). Of the sixty-nine words, twenty-three represent first layer bit-pattern words, twenty-three represent second-layer bit-significance words, and twenty-three re~resent third-layer bit-weight words.
Each set of twenty-three words correspond to the height of the character, and each sixteen-bit word correspond to the width of the character. (In Figure 7, for simplicity, eight bits rather than sixteen bits are used to depict the logical operations.) A listing of the pseudo code of the stored program which enables the system to perform the logical and other processing operations described above is presented below.
PSUEDO CODE
PROCEDURE: MAIN
Recognizes a codeline from a document image. -while (NOT END OF_DO~U.I~
{
if (FIND_FIRST _CHARACTER finds a - character) then RECOGNIZE_FIELD
else ; END OF DOCUMENT
.. }
PROCEDURE: FIND FIRST CHARACTER
do { ,:,~
SCAN_IMAGE_FOR_WHITE_TO_BLACK_TRANSITION
} .:
until (FOUND CHARACTER or END OF_DOCUMENT) ~
~: .
; ,:' , '~ '.
., ~ .
,., . ., . . . . - .,, ,. , ., , .. .,, ~, :..... ,.; . , : , : .
PROCEDURE: RECOGNIZE FIELD
; do : RECOGNIZE_CRiARACTER
move_to_next_position ; while (FOUND C~ARACTER) PROCEDURE: RECOGNIZE ~RiARACTER
. BINARIZE C~ARACTER_IMAGE
~: TEMPLATE_RECOGNITION_OF_CHARACTER
; if REJECTED
then ~ ENHANCED_BINARIZE_CHARACTER_IMAGE
TEMPLATE RECOGNITION_OF_CHARACTER
While the present invention is susceptible of embodiment in various forms, there is shown in the . drawings and described in the specification a presently preferred embodiment, with the understanding that the present disclosure is to be considered as an exemplification of the invention, and is not intended to limit the invention to the specific embodiment . illustrated. For example, in addition to the E13B
~ (and OCRA and OCRB) type characters discussed herein, the various calculations performed by the system can ~ be easily modified to enable the system to read images .. of any resolution with characters of various sizes.
~.
; ' ' ., , ~.
.. ' ' . , ., .
:
Background of the Invention This invention relates to the scanning of documents and the recognition of alphanumeric symbols generally, and particularly to optical scanning and recognition of alphanumeric characters such as E13B, OCRA, OCRB, and Farington 7B type fonts.
The E13B, OCRA, OCRB, and Farington 7B
character sets have received widespread use in financial transactions in the United States and abroad. Generally, the characters are recorded on documents such as bank drafts, checks, and other financial documents. For example, E13B type fonts are generally recorded in MICR ~magnetic ink character record) form for scanning by a magnetic scanner.
A vidicon camera system is sometimes used to scan the document and to record an image of it for archival and evidentiary purposes while, in a separate ~ operation, a magnetic scanner system is used to read ; the magnetic-ink characters (e.g., E13B characters).
This need for two separate scanner systems to perform two separate operations presents a costly, time-consuming problem. Further, hardware for processing E13B, OCRA, OCRB and Farrington 7B type character is often complex and expensive, and unduly time-consuming in performing cbaracter recognition. Also, such hardware often binarizes the entire image, introducing noise into the binarized image and causing loss of image data, making the recognition process more dif f icult.
A way of providing for image scanning and character recognition that would eliminate such a two-system two-operation requirement, and that would significantly reduce noise and image-processing time, is needed and would be useful.
~'- .' '~ ' :~ .
. .. - .. .. .. , ...... . . ., . .- ,.. .- . j ~.. - , , -, , ,~ . .
,, ~ f~
Summary of the Invention Accordingly, a system and method for optical character recognition are provided which obviate the need for separate magnetic scanner systems and the performance of separate magnetic scanning operations, and reduces noise and image-processing time without causing loss of image data.
In a first aspect of the invention, a system (apparatus) is provided comprising: (a) a CCD scanner for optically scanning an alphanumeric character on a document, the character and document having different opacities, and producing a plurality of gray-scale pixel values corresponding to said opacities, (b) a memory disposed for storing the gray-scale pixel values, a program, and a plurality of predetermined binary patterns, one of said patterns representing the character, and (c) a processor coupled to the scanner ~.
and to the memory, and operating under control of the program, for locating the character on the document by selecting from said gray-scale pixel values a matrix of values representative of the character, for converting the matrix values to binary data, and for selecting from the plurality of predetermined binary patterns a pattern that matches the converted binary data, said matching pattern serving to identify the character.
The CCD scanner includes a control logic, the processor includes a microprocessor, and the memory includes a ROM (read only memory) with a boot routine and a RAM (random access memory) with the program, the pixel values, and the binary patterns.
In a second aspect of the invention, a method is provided for optically identifying an alphanumeric character on a document, the character ~.
and document having different opacities, the character -;.
being represented by one of a plurality of :
.
,, _ .
~ :; 8 predetermined binary patterns, the method comprising the steps of: (a) optically scanning the character on the document and producing a plurality of gray-scale pixel values corresponding to said opacities, (b) locating the character by selecting, from said gray-scale pixel values, a matrix of values representative of the character, (c) converting the matrix values to binary data, and (d) selecting from the plurality of predetermined binary patterns a pattern that matches the converted binary data, said matching pattern identifying the character.
The characters are recorded on the document in, for example, E13B format. The camera (CCD
scanner) scans the document vertically, from bottom to top, starting at the bottom right corner of the document. The document is moved to the right continuously (with the width of each vertical scan corresponding to the diameter of one pixel) until the top left corner of the document is reached indicating that scanning has been completed.
In locating a character (i.e., the gray-scale pixel values representing the character) for proce~sing, the system positions a window over selected pixel values of the character (thereby framing the character) such that the sum of the selected pixel values within the window is a maximum.
The system then determines the arithmetic mean of the dark pixels within the window. It does this by determining the arithmetic mean of pre-selected ones of the selected pixel values (the preselected ones representing pixel values above a predetermined threshold, the predetermined threshold representing a value greater than the character's background level (determined by pre-scan of the character). The system then converts the selected pixel values to binary data relative to said mean (i.e., converts to one or zero depending on whether the magnitude of each selected :
pixel value is less than, or is greater than or equal to, said mean). The binary data is then logically compared to each of the predetermined binary patterns until a matching pattern is found which best identifies the character. The predetermined binary patterns represent binary templates, each pattern (template) having three bit configurations (levels), a first bit configuration representing the actual bit pattern of the character, a second bit configuration representing the significance of the bits of the actual bit pattern, and a third bit configuratlon representing weighted values for the bits of the actual bit pattern. The extent to which the converted binary data logically compares to each of the three bit configurations of each of the predetermined bit patterns is expressed as mismatch words, from which mismatch counts are calculated. The mismatch counts ;-are then compared to at least one predetermined threshold value, and the lowest mismatch count that is below (less than) the threshold value i8 selected as representative of the character.
Thus, as indicated above, the system performs segmentation of characters from the gray-scale image and binarizes the segmented characters for subsequent recognition, rather than binarizing the entire image for recognition of characters as do prior-art systems. Also, whereas such prior binarization of the entire image introduced undue amounts of error (noise) into the scanned data, the method of the present invention reduces such errors by using gray-scale values in the character-segmentation process. Further, the system is able to recognize different fonts ~i.e., different characters at different resolutions) merely by storing template tables for that font. -. ''-'' .
: .:
..
:~ .
... . - . --~: ~. . . . . .
. . - .. . . ., . , .. ,, . .: i ,.. . ... . . .. . .
- 5 - ~ ,t'l~.
Brief Description of the Drawings Figure 1 is a block diagram of the system of the present invention;
Figure 2 is a graphic illustration of the appearance of a character scanable and identifiable by the system of Figure l;
Figure 3 is a graphic illustration of a portion of a document with various characters recorded thereon;
Figure 4 is a graphic illustration of a scanned character having a matrix of pixels of different opacities (gray-scale values);
Figure 5 is a graphic illustration of a window used for framing a character;
Figure 6 iB a graphic illustration of the window of Figure 5 used in framing the character of Figure 4;
Figure 7 is a block diagram showing the results of various logical operations performed on binary data representing the scanned character;
Figure 8 is a block diagram of a matrix repre~enting various portions of the image of a framed character; and Figure 9 is a graphic/tabular illustration .-of a template and associated binary values used by the system of Figure 1.
Description of the Preferred Embodiment Referring now to Fig. 1, there is shown a system (apparatus) 11 of the present invention. The system includes and optical reader or scanner 13 for : :
scanning a document 19, and a computer system 21 for proce~sing data received from the scanner 13. The sC~nner may be a camera with a CCD (charge coupled device) 15 and a controller 17. The document 19 may be a bank draft or check with numeric character or .
.; .
other alphanumeric symbol 20 recorded thereon. The computer system 21 comprises a microprocessor 23, a ROM (read-only-memory) 25 with a boot program, and a RAM (random access memory) 27, with a stored program, for storing data from the scanning device and from the microprocessor and for storing predefined data.
The scanning operation is commenced at the bottom right corner 33 of the document l9 and ends at the top left corner 35. The scanner 13 scans the document l9 vertically, from bottom-to-top, as shown by arrow 37, first along right-most virtual scan line 39, then again from bottom-to-top along next-right-most virtual scan line 41, and so on until the top left corner of the document is reached. The virtual scan lines are arranged in parallel. The document 19 is moved continuously from left to right, as shown by arrow 43, while the scanner scans along successive scan lines, the width of each scan line corresponding to the diameter of a pixel (approximately .0127 cm or .005 inch).
An example of a read line of a document i9 shown in Figure 3. When a document is scanned, the pixel values generated by the scanner to represent the document, including the characters thereon, are stored in memory 27 (Figure 1). The characters are each stored as a data array (matrix) having, for example, typically one hundred rows, corresponding to the read line's height, and twelve hundred columns corresponding to the document'~ length (twelve hundred being the typical number of columns for a six-inch document at 200 pixels per inch). Within the data array, a character (when one is located and segmented as described below) is represented by a matrix of rows (e.g., twenty-one, corresponding to the character's height) and columns (e.g., fourteen, corresponding to the character's width). The character "three" shown in Figure 2, for example, may be represented in pixel . , .
. ~ .
array form as depicted in Figure 4. Each pixel 49 represents a portion of the document and has an opacity index or gray-scale value in the ran~e 0-255.
The dark pixels generally have a greater gray-scale value (approaching 255), and the light pixels generally have a lesser gray-scale value (approaching 0). Since characters are normally printed in a dark color compared to the generally light colored background of the document, the difference in opacity between the character and its background is easily determinable.
As indicated above, these gray-scale data values are generated by the scanner, upon scanning the document. The generated data are then stored in memory 27 (Figure 1) for processing by the system 21.
The system, under control of the stored program, examines the stored data values, and proceeds to locate a first character, then successive characters of a first field, then other characters of successive fields. Upon locating a character, the system proceeds to frame or delimit the character (i.e., to separate-out or segment the character from other characters). Starting with the stored gray-scale data values obtained from scanning the document from bottom-to-top along the right-most scan line 39 (Figures 1 and 2), and then from bottom-to-top along the next-right-most scan line 41, and so on, the system locates and frames (segments) each of the characters as described below. It does this by first searching the stored data for a rise in gray-scale value representing a transition from a light pixel to a dark pixel, vertically from bottom to top along the scan line. If a rise of more than, for example, forty the particular value, such as forty, being font dependent) is found between two gray-scale values of adjacent pixels in the scan direction 37, a dark pixel is deemed to be found. The region (of the document) proximate to the dark pixel is regarded as a region where a character is likely to be found. When a dark pixel is found, the system checks the character region (i.e., checks the stored data proximate to the dark pixel value) to determine if the dark pixel is part of a character. It checks for the presence of a character by establishing a segmentation window of sufficient size to effectively frame the character on the document, the position of the dark pixel corresponding to the center of the window. As depicted in Figure S, a segmentation window 53 is shown having left and right edges (columns) 55,57, and top and bottom edges (rows) 59,61. Bordering the left and right edges are columns (left+l) 63 and (right-l) 65, and bordering the top and bottom edges are rows (top+l) 67 and (bottom-l) 69.
As explained below with reference to Figures S and 6, a character is considered framed and positioned in the center of the window when the sum (Sw) of all the gray-scale pixel values in the window is maximum. To determine whether the sum (Sw) increases when the window is moved, the sums of the pixel values in each of the left and right columns 55,57 and in each of their bordering columns 63,65 are calculated. Likewise, the sums of the pixel values in each of the top and bottom columns 59,61 and in each of their bordering columns 67,69 are calculated. If, -for example, the sum in column 63 is found to be greater than the sum in column 57, and would cause the sum (Sw) in the window to increase when the window 53 is moved by one pixel leftward to include column 63 and exclude column 57, then the window is moved leftward. Likewise, if by moving the window by one pixel upward to include row 67 and exclude row 61, the sum (Sw) in the window would increase, then the window is moved upward. Thus, from the sum (cumulative count) of the pixel values of the respective rows and .
' h, ' . , . ~ . . . , ,. ' . .. ' . . ,, '' .. ' . , . ' . ' ', ~ , : : ' ' . :
columns adjoining (bordering) the window, relative to the sum of the pixel values of the respective rows and columns at the edges of the window, the direction can be determined in which the window can be moved to increase the sum (sw) of the pixel values within the window. As shown in Figure 6, according to the following counts:
Right-l column count = 0 Right column count = 0 Left column count = 100 Left+l column count = 100 Bottom-l row count = 0 Bottom row count = 0 Top row count = 170 Top+l row count = 170 the window count is increased by 100 upon a move leftward, and increased by 170 upon a move upward.
Therefore, the window is moved leftward and upward.
Thus, the window is moved in the direction which increases Sw, and this operation is repeated until no movement of the window in any of the four directions (left, right, up, down) increases Sw. The window is then considered to have framed (delimited/segmented) a character, with the character positioned at the center of the window. Thus, by means of the window a matrix of pixel values most representative of the character is selected (framed).
As a check to determine if there is enough information (dark-pixel gray-scale values) in the window to represent the framed character, the total number of pixels in the window that exceeds a predetermined confirmation threshold (e.g., a threshold gray-scale value of 100) is obtained. If the total number i~ greater than, for example, sixty (for the E13B font), then a character is confirmed to be found in the window. Otherwise, the object framed in the window is interpreted as noise, and the system proceeds to find the next character.
,, .
, - ::, ~ . , . . ,, . , , , . . . " . . , , .. "- . . - , ~ .
~ , , , . ~ . . , . . - . . . . .. " , , ., . , ... . - - . .
., .. .~, ,., . . . : .. .. . .. ..
,f ~
: . , After segmentation, the segmented (framed) character is scaled in preparation for identification (recognition). Sixteen bit words are used in the recognition process (i.e., width of template is 16 pixels). Because many characters are wider than 16 pixels (including E13B characters which are 19 pixels wide), they are converted to 16 bits after segmentation. This is accomplished by eliminating predetermined columns e.g., for the E13B font, columns 4, 9 & 14 are eliminated. (OCRA and OCRB fonts are 16 pixels wide, at two hundred pixels per inch, and do not need to have columns eliminated.) ~ The preceding describes how a character is ; located in preparation for processing, i.e., how it is effectively located on a document and framed (segmented) for later processing. The following describes how the framed character is converted to binary form then matched to a template (predetermined binary pattern) for the purpose of identifying (classifying/recognizing) the character.
In the conversion operation, the stored ;
gray-scale pixel values for each row of the framed character are converted to binary form. First a reference threshold value i~ calculated as the ~;
arithmetic mean of the dark pixels of the framed character (as previously described). Then, the gray- ~-scale pixel values are converted to binary form relative to the reference threshold. For example, if the calculated reference threshold is one hundred, then, using the binary equivalence scale ~hown below:
. . .
Gray-Scale Pixel Values Binary Equivalent !'' > 100 1 ( black) < 100 0 (white) a row of gray scale values of 80, 120, 130, 90, 85, 70, 110, 135 may be converted to binary data having the bit pattern 01100011, as shown in the eight-bit word 71 o~ Figure 7. , ' This binary data (which represents a row or slice of the framed character) and other rows of converted binary data representing the character are then compared to a plurality of templates, each template corresponding to a character. Each template represents a predetermined pattern of binary data and comprises three distinct layers (bit configurations).
The three eight-bit words 73, 75, 77 shown in Figure 7 represent an eight-bit row in each of the three layers. The first layer (bit-pattern layer, corresponding to pattern word 73) represents the actual black/white (0/1) bit pattern of the character.
The second layer (bit-significance layer, corresponding to bit-significance word 75) identifies the bit positions of the character that are significant and those that are not significant. One-bits identify significant bit positions where the bit pattern of the character is expected to remain unchanged regardless of small changes in size and shape of the character. Zero-bits identify non-significant bit positions where the bit pattern need not be the same for different sizes or shapes of the character. The third layer (bit-weight layer, corresponding to weight word 77) identifies which bits are critical, and hence have greater weight than others, in distinguishing between like (substantially similar) characters. In this third layer, l-bits are designated as weighted bits, and 0-bits are designated as non-weighted bits. For example, l-bits in the top row of the third layer of the template for the characters "O" and "U" would be useful in distinguishing between the two characters, and l-bits in the bottom row of the third layer of the templates for the characters "Q" and "O" would be useful in distinguishing between the Q and the O.
; As shown in Figure 7, various logic operations are performed by processor 23 (Figure 1) on - . . ... . .
; ~
the binary data 71 in order to identify the character.
(The logic operations are actually performed as sixteen-bit operations. However, for simplicicity, eight-bit operations are shown). First, the binary data is EXCLUSIVE-OR'd with the corresponding word 73 of the bit-pattern layer to produce a pattern of mismatch 74 between the binary data 71 and the predetermined correct pattern 73. This mismatch pattern 74 is then logically ANDed with the corresponding word 75 of the bit-significance layer for producing a mismatch word 76, representing the extent of mismatch independent of character size and shape. The mismatch word 76 is then weighted, i.e., logically ANDed with the corresponding word 77 in the bit-weight layer to produce a weighted mismatched indicator 78. (If weighting word 77 contained no 1-bits, no weighting or ANDing would have been performed (in order to save processing time since the outcome would be zero), and the system would have proceeded -with processing the next row of binary data.) Following the weighting operation, a mismatch count (MCR) is calculated for the row of binary data 71. This calculation is performed by processor ~3 (Figure 1) as follows:
MCR = MCw + (WOC x WF) (1) where, MCW represents a mismatch count for the mismatchword (i.e., the number of ones -:
found in mismatch word 76), WOC represents a mismatch indicator or weight out-put count (i.e., the number of ones found in mismatch indicator'!' ' word 78), and WF represents a predetermined weighing factor (e.g., the integer 2 for E13B
type characters).
.:
., :
.~ .
Thus, for the mismatch bit pattern shown in the words 76 and 78 in Figure 7, it can be seen that MCW = 2, WOC = 1, and WF = 2. Therefore, the mismatch count (MCR) calculated for the row of binary data 71 would equal four (i.e., 2+(1x2) according to equation 1 above).
After calculating the mismatch count for one row, the system proceeds to calculate, as indicated above, the mismatch count for all remaining rows of the character. The mismatch counts for all the rows of the character are then summed to produce a template mismatch count (i.e., a mismatch count for the template as applied to the framed character). In a similar manner, for the other templates of the character set for the font being processed, a mismatch count is produced for each template relative to the framed character. The two lowest template mismatch counts and their associated template numbers are stored in memory as the various templates are processed. The criterion for identifying a character is: if the template with the lowest count is below a threshold (a reje~t threshold of 40 for E13B fonts) and is lower than the next lowest template by a predetermined amount ~a difference threshold of 5 for El3~ fonts), it identifies the character. ~ -Template recognition, as described previously, can be performed in a total of nine positions as shown in Figure 8. This is done in order to identify characters that may not be perfectly framed due to factors like noise in the image. In the example shown in Figure 8, the template and input (binary data) pattern sizes are 23 rows (high) by 16 columns (wide). The three layers of the template (pattern layer, bit significance layer and weighting layer) are all shifted (moved) the same amount and in the same direction. In the description of shift positions presented below, template row refer~ to the ~-'~ c~
bit pattern row, bit significance row and bit weight row.
In the center position 87, the template 96 is placed directly over the input pattern 98. Only rows 2 through 22 are used in this match. Thus, template row l is compared to pattern row 1, template row two is compared to pattern row 2 and so on. All columns (l through 16) are used in the center horizontal positions 81, 87, 93. In the upper center position 81, rows l through 22 of the template are compared to rows 2 through 23 of the input pattern.
This is the same as shifting the template pattern vertically upwards by one row. In this case, template row l is compared to input pattern row 2, template row 2 i9 compared to input pattern row 3 etc. All columns of the template and input pattern are compared in this situation (the horizontal center). Similarly, in the lower center position 93, the template is shifted down by one row, thus matching template rows 2 through 23 against input pattern rows l through 22.
The horizontal shift positions correspond to :
a shifting of the template 96 to the left or right over the input pattern 98. In the center left position 85, template columns l through 15 are used ;-and input pattern columns 2 through 16 are used.
(Since it is a center position vertically, rows 2 through 22 are used for both the input pattern and the template.) Thus, template word bit l is compared to pattern word bit 2, template word bit 3 is compared to pattern word bit 4, and so on. ~ -. . .
For example, if the template pattern layer word is: ' 0011111100001111 .: ' and the input character pattern word is:
0010111100011110 ~:'.'' The template is shifted left l bit (column):
0011111100001111 ~:
' .
.~, . .
' ,":-and the result of the exclusive Or operation would be:
It should be noted that when a shift left operation is performed, the rightmost bit (least significant bit, LSB) is 0. Thus, the LSB of the bit-significant word will be 0 also (since all three layers are shifted in the same direction and amount) and, as such, the LSB
of the mismatch word 76 (after the AND operation with the bit-significant word) will always be 0.
Similarly, a shift to the right makes the leftmost bit (most significant bit, MSB) a 0 and, as such, the MSB
of the mismatch word 76 (after the AND operation) will always be 0.
A horizontal shift to the right (position 89) is similar to the left shift, only in the opposite direction. Thus, bit 2 of the template word would line up (be compared to) bit 1 of the input pattern word.
The corner positions (positions 95, 91, 83, 79) represent a combination of a vertical shift of one row and a horziontal shift of one column. A~ an example, in the upper left position 79, the template row 1 is shifted left by one bit position and compared to pattern word 2, and template row 2 is shifted left by one bit position and compared to pattern word 3, and so on.
To recognize a character using this method requires nine passes per template times the number of templates in the character set. For example, an alphanumeric character set with 49 characters would require a total of 441 (9*49) matches being done.
Thi9 can be too time consuming. To promote faster character identification, only the center position of each template is compared initially to the input pattern. The lowest two mismatch counts and their corresponding template numbers are stored in memory 27 ~-.
.. .. .
as the various templates are processed. At the end of this first pass, the other eight positions of only these two templates are processed to find the best possible (i.e., lowest) total mismatch against the input pattern. Thu~, only 6S (1*49 + 2*8) matches need to be made ordinarily (i.e., about 15% of the 441 matches that were required).
A description of the criteria for accepting or rejecting a character i5 presented below.
The lowest template mismatch count is compared to a predetermined reject threshold for the character that the template represents. If the template mismatch count for this character is below this threshold and the difference in template mismatch counts for the lowest two templates (second lowest mismatch minus lowest mi~match) is greater than a predetermined amount (e.g., 10), the template with the lowest template mismatch identifies the framed ;
character. The reason for checking if the mismatch counts are too close is done to discriminate between characters that are very much alike (e.g., Q and 0).
If the mismatches are similar, the character is rejected rather than risking a misidentification (misread). If no template mismatch falls below the reject threshold discussed above or the two template mismatch counts are too close together, a post processing operation (described below) is performed.
If after the post-processing operation the framed character is still rejected, its image is enhanced and converted to binary form (as described below). The recognition process (described above) i~ then performed again. If the character i~ identified (satisfies the above criteria), the system then proceeds to process (locate and frame) the next character as described above.
By examining the center positions only, the pos~ibility may be missed that a character, other than ,:, ., , , , . ,; .. . .. . . . . . . . . .
the closest two, may have a closer count in a position other than the center and be the correct choice. The post processing operation can remedy such a situation.
If the lowest two characters chosen are both over the predetermined reject threshold for that individual character or the inter-character distance (second lowest mismatch-lowest mismatch) is too small, a number of other templates may be tested in the other eight positions in an attempt to find a better match.
The choice of which templates to process is made, for example, based on the closest character's similarity to the templates. For example, the closest two characters (those with the lowest mis~atch counts) chosen, the letters D and O, both have mismatch counts above their individual reject threshold. In post processing operation, a predetermined set of characters that are similar to D and O would then be processed. This could include characters 0 (zero~, Q, C and U. The mismatch counts of the closest two characters after this operation are checked (the one with the lowest mismatch count is checked first) to see if they meet the reject criterion and inter-character distance criterion. If at least one character meets the criteria, that character is returned as the result, and processing (locating and framing) of the next character in the image proceeds.
If neither of the above characters meets the reject criteria, or the inter-character distance is too small, the framed character is rejected. In this case, its image is enhanced, (as described below), the character is rebinarized, and the recognition process described above is repeated a second time. If ~he character is recognized, it is returned. Otherwis~, a reject character code is returned as the result, and processing (locating and framing) of the next character on the document continùes.
., .
In addition to the reject threshold described above, an exit threshold (e.g., the value 10) may also be used in the character identification process as described below. To enhance processing speed, successive template mismatch counts are compared to the exit threshold value, and the first template with a template mismatch count that fall~ :
below the exit threshold value would identify the character. In which event, the remaining templates are not processed (which enhances processing speed).
Further, to speed the character identification process, if, while the individual rows of a template are being processed, the mismatch count of the current template (i.e., the template currently being processed) is found to be larger than the second-lowest-count-obtained-thus-far, processing of the current template may be discontinued since this current template is no longer a contender to be one of the two with the lowest template mismatch counts. To further enhance recognition speed, the templates may be presequenced in the order in which they are most likely to occur (e.~., numbers before vowels followed by consonants in an alphanumeric font). If the mismatch is below the exit threshold (as mentioned above), the remainder of the templates need not be examined. Thus those characters occurring most frequently should be examined first. Also, to promote faster character identification, only the center position may be tested by the templates, after which, if the lowest template mismatch count is above the exit threshold, the remaining eight positions for each of the two template~ closest to the exit threshold may be processed, and the template with the lowest template mismatch count selected a~ representative of the character. Further, to speed identification, the remaining eight positions (Figure 8) may be processed in the following sequence 79, 81, 83, 85, 89, 91, 93, 95. This is the order in which the character is most likely to be positioned.
After identifying the framed character, the system proceeds to process the next character (i.e., to process the next group of stored pixel values representing the region on the document proximate to the framed character) by moving the segmentation window a selected number of pixels (e.g., three pixels) to the left of the previously framed character. The character in this proximate region is then framed (segmented) and identified as described above. All of the remaining characters (remaining stored pixel data) are processed in this manner until end-of-field is detected (i.e., no more characters have been located or an end-of-field character has been located).
If, during the character-identification process a character i9 rejected, its image is enhanced to facilitate identification. The image may be enhanced, for example, by producing a character image having constant stroke widths (i.e., having a constant number of dark pixels across the width of each vertical stroke or horizontal stroke of the character). This is accomplished as described below.
First, the gray-scale pix~l values (e.g., 6-bit, 8-bit, or 16-bit values) representing the character are converted to a lesser range of values (e.g., 3-bit gray-scale pixel values for faster processing), then the individual pixels of the character image are examined and only those pixels that are the darkest two or three pixels of the stroke are turned black, others are turned white. This produced a binary image with constant stroke widths as described for example in U.S. Patent No. 4,625,330, incorporated by reference herein).
As an assist for improving character identification, the edges of each character may be : ' .
processed as described below. As mentioned above, each template used in the character identification process corresponds to one character of the selected -character set (e.g., the E13B character set) and, as shown in Figures 2 and 4, each character is fourteen pixels wide by twenty-one pixels high. One extra column is added to the left and to the right of the character, and one extra row is added to the top and to the bottom of the character. This results in a character size of 16 by 23. Corresponding to each character, each template is sixteen pixels wide by twenty-three pixels high. Each template has an extra column on each side (right side and left side) o~ the fourteen-column character, and one extra row on the top and an extra row on the bottom of the twenty-one-row character. Each extra row comprises sixteen white pixels, and each extra column comprises twenty-three white pixels. The extra rows and columns serve to surround the 14 x 21 character region with a border of white pixels for better identification of the edge portions of certain characters, as explained below.
For example, if the character "E" is matched against a 14 x 21 "F" template, and the template is shifted upward one pixel position, the pixel values at the bottom of the "E" would be lost since it could not be used in generating the template mismatch count. This could result in mis-identification of the E as an F
since the mismatch count would be low. With the 16 x 23 template, the bottom of the E would not be lost and would contribute substantially to the mismatch count against (relative to) the bottom white row of the F
template, thereby avoiding mis-identification of the E
as an F. A similar explanation holds with respect to the right and left column positions of the 16 x 23 template, the columns serve to reduce mis-identification between characters such as "B" and "3".
., .
:
.
. . . .. . . . ~ -- . . - ..
As shown in Figure 9, each 16 x 23 template may be represented by sixty-nine sixteen-bit words of RAM 27 (corresponding to a scan of two hundred pixels per inch). Of the sixty-nine words, twenty-three represent first layer bit-pattern words, twenty-three represent second-layer bit-significance words, and twenty-three re~resent third-layer bit-weight words.
Each set of twenty-three words correspond to the height of the character, and each sixteen-bit word correspond to the width of the character. (In Figure 7, for simplicity, eight bits rather than sixteen bits are used to depict the logical operations.) A listing of the pseudo code of the stored program which enables the system to perform the logical and other processing operations described above is presented below.
PSUEDO CODE
PROCEDURE: MAIN
Recognizes a codeline from a document image. -while (NOT END OF_DO~U.I~
{
if (FIND_FIRST _CHARACTER finds a - character) then RECOGNIZE_FIELD
else ; END OF DOCUMENT
.. }
PROCEDURE: FIND FIRST CHARACTER
do { ,:,~
SCAN_IMAGE_FOR_WHITE_TO_BLACK_TRANSITION
} .:
until (FOUND CHARACTER or END OF_DOCUMENT) ~
~: .
; ,:' , '~ '.
., ~ .
,., . ., . . . . - .,, ,. , ., , .. .,, ~, :..... ,.; . , : , : .
PROCEDURE: RECOGNIZE FIELD
; do : RECOGNIZE_CRiARACTER
move_to_next_position ; while (FOUND C~ARACTER) PROCEDURE: RECOGNIZE ~RiARACTER
. BINARIZE C~ARACTER_IMAGE
~: TEMPLATE_RECOGNITION_OF_CHARACTER
; if REJECTED
then ~ ENHANCED_BINARIZE_CHARACTER_IMAGE
TEMPLATE RECOGNITION_OF_CHARACTER
While the present invention is susceptible of embodiment in various forms, there is shown in the . drawings and described in the specification a presently preferred embodiment, with the understanding that the present disclosure is to be considered as an exemplification of the invention, and is not intended to limit the invention to the specific embodiment . illustrated. For example, in addition to the E13B
~ (and OCRA and OCRB) type characters discussed herein, the various calculations performed by the system can ~ be easily modified to enable the system to read images .. of any resolution with characters of various sizes.
~.
; ' ' ., , ~.
.. ' ' . , ., .
Claims (37)
1. Optical character recognition system comprising:
means for optically scanning a character recorded on a document, the character and document having different opacities, and producing a plurality of gray-scale values corresponding to said opacities, said gray-scale values representing light and dark pixels;
memory means disposed for storing the gray-scale values, said memory means including a stored program, and predetermined data including a plurality of predetermined binary patterns, one of said predetermined binary patterns representing the character;
processing means coupled to the scanning means and to the memory means, and operating under control of the stored program, for locating the character on the document by selecting from said gray-scale values a matrix of values representative of the character, for converting the matrix values to binary data, and for selecting from the plurality of predetermined binary patterns a pattern that matches the converted binary data, said matching pattern serving to identify the character.
means for optically scanning a character recorded on a document, the character and document having different opacities, and producing a plurality of gray-scale values corresponding to said opacities, said gray-scale values representing light and dark pixels;
memory means disposed for storing the gray-scale values, said memory means including a stored program, and predetermined data including a plurality of predetermined binary patterns, one of said predetermined binary patterns representing the character;
processing means coupled to the scanning means and to the memory means, and operating under control of the stored program, for locating the character on the document by selecting from said gray-scale values a matrix of values representative of the character, for converting the matrix values to binary data, and for selecting from the plurality of predetermined binary patterns a pattern that matches the converted binary data, said matching pattern serving to identify the character.
2. The system as in claim 1 wherein said character has a right, left, top and bottom edge, and said scanning means includes a means for reading from the bottom edge to the top edge in a vertical direction.
3. The system as in claim 2 wherein said processing means includes a means for framing a region having a plurality of pixels proximate to a dark pixel.
4. The system as in claim 3 wherein said framing means includes a means for defining a window for bordering said region.
5. The system as in claim 4 wherein said framing means includes a means for positioning the window so that the sum of the gray-scale values within the window is a maximum, corresponding to the character being centrally located in the window, said maximum-sum values corresponding to a matrix of selected pixel values representative of the character.
6. The system as in claim 5 wherein said framing means also includes a means for confirming the presence of a character within the window, by determining the total number of pixels in the window that exceeds a predetermined confirmation threshold.
7. The system as in claim 6 wherein said processing means includes a means for calculating an average threshold value for said selected pixel values.
8. The system as in claim 7 wherein said calculating means includes a means for producing binary data representative of the magnitudes of said selected pixel values relative to said average threshold value.
9. The system as in claim 8 wherein said processing means includes a means for producing a mismatch count for each predetermined binary pattern said count being indicative of the extent of mismatch between the predetermined binary pattern and the converted binary data.
10. The system as in claim 9 wherein said means for producing a mismatch count includes a means for selecting from the mismatch counts the count that best satisfies at least one predetermined threshold criterion.
11. The system as in claim 10 wherein the means for producing a mismatch count includes a means for producing mismatch words and calculating a count based on the contents of selected ones of said words, said words being indicative of the extent of mismatch between the converted binary data and selected bit configurations of said predetermined binary pattern.
12. The system as in claim 11 wherein each predetermined binary pattern includes first, second, and third bit configurations, said first bit configuration representing an actual bit pattern of the character, and said means for producing a mismatch count includes a logic means for performing an exclusive-OR operation between the converted binary data and the first bit configuration to produce a first mismatch word.
13. The system as in claim 12 wherein said second bit configuration represents predetermined significance of respective bits of the actual bit pattern, and said means for producing a mismatch count includes a logic means for performing an AND operation between the first mismatch word and the second bit configuration to produce a second mismatch word.
14. The system as in claim 13 wherein said third bit configuration represents weighted values of respective bits of the actual bit pattern, and said means for producing a mismatch count includes a logic means for performing an AND operation between the second mismatch word and the third bit configuration to produce a third mismatch word.
15. The system as in claim 14 wherein each of said predetermined binary patterns represents a predetermined template, and said framed character may be compared against a predetermined template in each of nine positions, a center position and eight non-center positions, and said means for producing a mismatch count includes a means for positioning a template relative to the center portion of the framed character.
16. The system as in claim 15 wherein said means for positioning places the predetermined binary pattern relative to one of the non-center portions of the framed character.
17. The system as in claim 16 wherein said gray-scale values are in the range 0 to 255.
18. The system as in claim 17 wherein said character is a member of a selected character set.
19. The system as in claim 18 wherein said character and said window are each of a predetermined size for a selected font.
20. The system as in claim 19 wherein each bit-configuration of said predetermined binary pattern is a first predetermined number of pixels wide by a second predetermined number of pixels high.
21. Method of optically identifying a character on a document, the character and document having different opacities, the character being represented by one of a plurality of predetermined binary patterns, the method comprising the steps of:
optically scanning the document and producing a plurality of gray-scale values corresponding to said opacities, said gray-scale values representing light and dark pixels;
locating the character by selecting, from said gray-scale values, a matrix of values representative of the character;
converting the matrix values to binary data; and selecting from the plurality of predetermined binary patterns a pattern that matches the converted binary data, said matching pattern serving to identify the character.
optically scanning the document and producing a plurality of gray-scale values corresponding to said opacities, said gray-scale values representing light and dark pixels;
locating the character by selecting, from said gray-scale values, a matrix of values representative of the character;
converting the matrix values to binary data; and selecting from the plurality of predetermined binary patterns a pattern that matches the converted binary data, said matching pattern serving to identify the character.
22. The method of claim 21 where the step of optically scanning the document includes the step of reading each character on the document from bottom to top in a vertical direction.
23. The method of claim 22 where the step of locating the character includes the step of framing a region having a plurality of pixels proximate to a dark pixel.
24. The method of claim 23 where the step of framing includes the step of defining a window for bordering said region.
25. The method of claim 24 where the step of framing also includes the step of positioning the window so that the sum of the gray-scale values within the window is a maximum, corresponding to the character being centrally located in the window, said maximum-sum values corresponding to a matrix of selected pixel values representative of the character.
26. The method of claim 25 where the step of framing the character includes the step of confirming the presence of a character within the window, by determining the total number of pixels in the window that exceeds a predetermined confirmation threshold.
27. The method of claim 26 where the step of converting includes the step of calculating an average threshold value from the values represented by the dark pixels of the framed character.
28. The method of claim 27 where the step of converting also includes the step of producing binary data representative of the magnitudes of said selected pixel values relative to said average threshold value.
29. The method of claim 28 where the step of selecting a predetermined binary pattern that matches the converted binary data includes the step of producing a mismatch count for each predetermined binary pattern, said count being indicative of the extent of mismatch between the predetermined binary pattern and the converted binary data.
30. The method of claim 29 where the step of selecting a predetermined binary pattern also includes the step of selecting from the mismatch counts the count that best satisfies at least one predetermined threshold criterion.
31. The method of claim 30 where the step of producing a mismatch count includes the steps of producing mismatch words and calculating a count based on the contents of selected ones of said words, said words being indicative of the extent of mismatch between the converted binary data and selected bit configurations of said predetermined binary pattern.
32. The method of claim 31 where each predetermined binary pattern includes first, second, and third bit configurations, said first bit configuration representing an actual bit pattern of the character, and the step of producing a mismatch count includes the step of performing an exclusive-OR
logical operation between the converted binary data and the first bit configuration for producing a first mismatch word.
logical operation between the converted binary data and the first bit configuration for producing a first mismatch word.
33. The method of claim 32 where said second bit configuration represents predetermined significance of respective bits of the actual bit pattern, and the step of producing a mismatch count includes the step of performing an AND logical operation between the first mismatch word and the second bit configuration for producing a second mismatch word.
34. The method of claim 33 where said third bit configuration represents weight values of respective bits of the actual bit pattern, and the step of producing a mismatch count further includes the step of performing AND logical operation between the second mismatch word and the third bit configuration for producing a third mismatch word.
35. The method of claim 34 where the framed character may be divided virtually into a three-by-three matrix representing nine portions, a center portion and eight non-center portions, and the step of producing a mismatch count includes the step of positioning a predetermined binary pattern relative to the center portion of the framed character.
36. The method of claim 35 where the step of positioning includes placing the predetermined binary pattern relative to one of the eight non-center portions of the framed character.
37. Method of identifying a character imaged on a document, the character and document having different opacities, the document being optically scanable, with images thereon representable as a plurality of stored gray-scale pixel values corresponding to said opacities, the character being represented by one of a plurality of predetermined binary patterns, the method comprising the steps of:
locating the character by selecting from said gray-scale pixel values a matrix of values representing the character;
converting the matrix values to binary data; and identifying the character by selecting, from the plurality of predetermined binary patterns, a pattern that matches the converted binary data.
locating the character by selecting from said gray-scale pixel values a matrix of values representing the character;
converting the matrix values to binary data; and identifying the character by selecting, from the plurality of predetermined binary patterns, a pattern that matches the converted binary data.
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| US07/635,899 US5091968A (en) | 1990-12-28 | 1990-12-28 | Optical character recognition system and method |
| US635,899 | 1990-12-28 |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CA2049758A1 CA2049758A1 (en) | 1992-06-29 |
| CA2049758C true CA2049758C (en) | 1998-05-19 |
Family
ID=24549574
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CA002049758A Expired - Fee Related CA2049758C (en) | 1990-12-28 | 1991-08-23 | Optical character recognition system and method |
Country Status (5)
| Country | Link |
|---|---|
| US (1) | US5091968A (en) |
| EP (1) | EP0492979B1 (en) |
| JP (1) | JPH04361391A (en) |
| CA (1) | CA2049758C (en) |
| DE (1) | DE69131006T2 (en) |
Families Citing this family (95)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6067379A (en) * | 1988-12-09 | 2000-05-23 | Cognex Corporation | Method and apparatus for locating patterns in an optical image |
| CA2049866C (en) * | 1990-12-28 | 1998-06-30 | Raymond L. Higgins | System and method for optical recognition of bar-coded characters |
| US5272766A (en) * | 1991-01-14 | 1993-12-21 | Ncr Corporation | OCR system for recognizing user-specified custom fonts in addition to standard fonts using three-layer templates |
| CA2052450C (en) * | 1991-01-14 | 1998-08-18 | Raymond L. Higgins | Ocr system for recognizing user-specified custom fonts in addition to standard fonts |
| US5299269A (en) * | 1991-12-20 | 1994-03-29 | Eastman Kodak Company | Character segmentation using an associative memory for optical character recognition |
| US5859923A (en) * | 1992-12-29 | 1999-01-12 | Cognex Corporation | Mark quality inspection apparatus and method |
| EP0677811A1 (en) * | 1994-04-15 | 1995-10-18 | Canon Kabushiki Kaisha | Image processing system with on-the-fly JPEG compression |
| US5802204A (en) * | 1995-06-07 | 1998-09-01 | Microchip Technology Incorporated | Handwritten character translator using fuzzy logic |
| US6026176A (en) | 1995-07-25 | 2000-02-15 | Cognex Corporation | Machine vision methods and articles of manufacture for ball grid array inspection |
| AUPN462995A0 (en) * | 1995-08-07 | 1995-08-31 | Mikoh Technology Limited | Optical memory data detector and authenticator |
| AU717222B2 (en) * | 1995-08-07 | 2000-03-23 | Mikoh Technology Limited | Optical image authenticator |
| US5729621A (en) * | 1995-08-31 | 1998-03-17 | Ncr Corporation | Method and apparatus for magnetic ink character recognition using a magneto-resistive read head |
| US5872870A (en) * | 1996-02-16 | 1999-02-16 | Cognex Corporation | Machine vision methods for identifying extrema of objects in rotated reference frames |
| US5909504A (en) * | 1996-03-15 | 1999-06-01 | Cognex Corporation | Method of testing a machine vision inspection system |
| US6298149B1 (en) | 1996-03-21 | 2001-10-02 | Cognex Corporation | Semiconductor device image inspection with contrast enhancement |
| US6259827B1 (en) | 1996-03-21 | 2001-07-10 | Cognex Corporation | Machine vision methods for enhancing the contrast between an object and its background using multiple on-axis images |
| US5978502A (en) * | 1996-04-01 | 1999-11-02 | Cognex Corporation | Machine vision methods for determining characteristics of three-dimensional objects |
| US5956468A (en) * | 1996-07-12 | 1999-09-21 | Seiko Epson Corporation | Document segmentation system |
| US6137893A (en) * | 1996-10-07 | 2000-10-24 | Cognex Corporation | Machine vision calibration targets and methods of determining their location and orientation in an image |
| US7207478B1 (en) | 1999-11-30 | 2007-04-24 | Diebold, Incorporated | Check accepting and cash dispensing automated banking machine system and method |
| US5960125A (en) | 1996-11-21 | 1999-09-28 | Cognex Corporation | Nonfeedback-based machine vision method for determining a calibration relationship between a camera and a moveable object |
| US5953130A (en) * | 1997-01-06 | 1999-09-14 | Cognex Corporation | Machine vision methods and apparatus for machine vision illumination of an object |
| US6075881A (en) * | 1997-03-18 | 2000-06-13 | Cognex Corporation | Machine vision methods for identifying collinear sets of points from an image |
| US5974169A (en) * | 1997-03-20 | 1999-10-26 | Cognex Corporation | Machine vision methods for determining characteristics of an object using boundary points and bounding regions |
| US6141033A (en) * | 1997-05-15 | 2000-10-31 | Cognex Corporation | Bandwidth reduction of multichannel images for machine vision |
| US6608647B1 (en) | 1997-06-24 | 2003-08-19 | Cognex Corporation | Methods and apparatus for charge coupled device image acquisition with independent integration and readout |
| US7519558B2 (en) | 1997-08-27 | 2009-04-14 | Ballard Claudio R | Biometrically enabled private secure information repository |
| US8024269B1 (en) | 1997-08-27 | 2011-09-20 | Datatreasury Corporation | Remote image capture with centralized processing and storage |
| US5910988A (en) * | 1997-08-27 | 1999-06-08 | Csp Holdings, Inc. | Remote image capture with centralized processing and storage |
| US5978080A (en) * | 1997-09-25 | 1999-11-02 | Cognex Corporation | Machine vision methods using feedback to determine an orientation, pixel width and pixel height of a field of view |
| US6025854A (en) * | 1997-12-31 | 2000-02-15 | Cognex Corporation | Method and apparatus for high speed image acquisition |
| US6282328B1 (en) | 1998-01-28 | 2001-08-28 | Cognex Corporation | Machine vision systems and methods for morphological transformation of an image with non-uniform offsets |
| US6236769B1 (en) | 1998-01-28 | 2001-05-22 | Cognex Corporation | Machine vision systems and methods for morphological transformation of an image with zero or other uniform offsets |
| US6381375B1 (en) | 1998-02-20 | 2002-04-30 | Cognex Corporation | Methods and apparatus for generating a projection of an image |
| US6215915B1 (en) | 1998-02-20 | 2001-04-10 | Cognex Corporation | Image processing methods and apparatus for separable, general affine transformation of an image |
| JP4338155B2 (en) * | 1998-06-12 | 2009-10-07 | キヤノン株式会社 | Image processing apparatus and method, and computer-readable memory |
| US6381366B1 (en) | 1998-12-18 | 2002-04-30 | Cognex Corporation | Machine vision methods and system for boundary point-based comparison of patterns and images |
| US6687402B1 (en) | 1998-12-18 | 2004-02-03 | Cognex Corporation | Machine vision methods and systems for boundary feature comparison of patterns and images |
| US7611048B1 (en) | 1999-11-30 | 2009-11-03 | Diebold, Incorporated | Check accepting and cash dispensing automated banking machine system and method |
| US6684402B1 (en) | 1999-12-01 | 2004-01-27 | Cognex Technology And Investment Corporation | Control methods and apparatus for coupling multiple image acquisition devices to a digital data processor |
| US6748104B1 (en) | 2000-03-24 | 2004-06-08 | Cognex Corporation | Methods and apparatus for machine vision inspection using single and multiple templates or patterns |
| US7386511B2 (en) * | 2000-04-28 | 2008-06-10 | Netdeposit Inc. | Methods and systems for processing financial instrument deposits |
| US7216106B1 (en) | 2000-04-28 | 2007-05-08 | Netdeposit, Inc. | Method and system for processing financial instrument deposits physically remote from a financial institution |
| US7181430B1 (en) | 2000-04-28 | 2007-02-20 | Netdeposit, Inc. | Method and system for processing financial instrument deposits physically remote from a financial institution |
| US7475040B2 (en) * | 2000-04-28 | 2009-01-06 | Netdeposit, Inc. | Return item early notification and return |
| US7006669B1 (en) | 2000-12-31 | 2006-02-28 | Cognex Corporation | Machine vision method and apparatus for thresholding images of non-uniform materials |
| US10453151B2 (en) | 2001-02-01 | 2019-10-22 | Kris Engineering, Inc. | Receipts scanner and financial organizer |
| US7746510B2 (en) | 2001-02-01 | 2010-06-29 | Pandipati Radha K C | Receipts scanner and financial organizer |
| US20030059099A1 (en) * | 2001-09-27 | 2003-03-27 | Longford Equipment International Limited | Optical character recognition system |
| US6873715B2 (en) * | 2002-01-31 | 2005-03-29 | You-Ti Kuo | System of central signature verifications and electronic receipt transmissions |
| US7295694B2 (en) * | 2002-02-22 | 2007-11-13 | International Business Machines Corporation | MICR-based optical character recognition system and method |
| US7979348B2 (en) * | 2002-04-23 | 2011-07-12 | Clearing House Payments Co Llc | Payment identification code and payment system using the same |
| KR100489550B1 (en) * | 2002-08-19 | 2005-05-16 | (주)신우아이티 | Apparatus for currency calculation which can extract serial number and method for the same |
| CA2502811C (en) * | 2002-10-21 | 2020-09-29 | Raphael Spero | System and method for capture, storage and processing of receipts and related data |
| US20040243489A1 (en) * | 2003-05-27 | 2004-12-02 | International Business Machines Corporation | Expense accounting data management based on electronic expense document |
| US7280684B2 (en) * | 2003-07-30 | 2007-10-09 | International Business Machines Corporation | Method and system for ongoing performance monitoring of a character recognition system |
| US20050097046A1 (en) | 2003-10-30 | 2005-05-05 | Singfield Joy S. | Wireless electronic check deposit scanning and cashing machine with web-based online account cash management computer application system |
| US8725607B2 (en) | 2004-01-30 | 2014-05-13 | The Clearing House Payments Company LLC | Electronic payment clearing and check image exchange systems and methods |
| US7386160B2 (en) * | 2004-12-02 | 2008-06-10 | International Business Machines Corporation | System and method for determining image resolution using MICR characters |
| US7639861B2 (en) | 2005-09-14 | 2009-12-29 | Cognex Technology And Investment Corporation | Method and apparatus for backlighting a wafer during alignment |
| US8111904B2 (en) * | 2005-10-07 | 2012-02-07 | Cognex Technology And Investment Corp. | Methods and apparatus for practical 3D vision system |
| US8162584B2 (en) * | 2006-08-23 | 2012-04-24 | Cognex Corporation | Method and apparatus for semiconductor wafer alignment |
| US8708227B1 (en) | 2006-10-31 | 2014-04-29 | United Services Automobile Association (Usaa) | Systems and methods for remote deposit of checks |
| US7873200B1 (en) | 2006-10-31 | 2011-01-18 | United Services Automobile Association (Usaa) | Systems and methods for remote deposit of checks |
| US10380559B1 (en) | 2007-03-15 | 2019-08-13 | United Services Automobile Association (Usaa) | Systems and methods for check representment prevention |
| US20080310721A1 (en) | 2007-06-14 | 2008-12-18 | John Jinhwan Yang | Method And Apparatus For Recognizing Characters In A Document Image |
| US9058512B1 (en) | 2007-09-28 | 2015-06-16 | United Services Automobile Association (Usaa) | Systems and methods for digital signature detection |
| US9159101B1 (en) * | 2007-10-23 | 2015-10-13 | United Services Automobile Association (Usaa) | Image processing |
| US20090198486A1 (en) * | 2008-02-05 | 2009-08-06 | National Tsing Hua University | Handheld electronic apparatus with translation function and translation method using the same |
| US10380562B1 (en) | 2008-02-07 | 2019-08-13 | United Services Automobile Association (Usaa) | Systems and methods for mobile deposit of negotiable instruments |
| JP2009199102A (en) * | 2008-02-19 | 2009-09-03 | Fujitsu Ltd | Character recognition program, character recognition device and character recognition method |
| US10504185B1 (en) | 2008-09-08 | 2019-12-10 | United Services Automobile Association (Usaa) | Systems and methods for live video financial deposit |
| DE102008042080A1 (en) * | 2008-09-15 | 2010-03-25 | Bundesdruckerei Gmbh | Automatic optical character capture method, computer program product, data processing system and scanner |
| US8452689B1 (en) | 2009-02-18 | 2013-05-28 | United Services Automobile Association (Usaa) | Systems and methods of check detection |
| US10956728B1 (en) | 2009-03-04 | 2021-03-23 | United Services Automobile Association (Usaa) | Systems and methods of check processing with background removal |
| US9779392B1 (en) | 2009-08-19 | 2017-10-03 | United Services Automobile Association (Usaa) | Apparatuses, methods and systems for a publishing and subscribing platform of depositing negotiable instruments |
| US8977571B1 (en) | 2009-08-21 | 2015-03-10 | United Services Automobile Association (Usaa) | Systems and methods for image monitoring of check during mobile deposit |
| US8699779B1 (en) | 2009-08-28 | 2014-04-15 | United Services Automobile Association (Usaa) | Systems and methods for alignment of check during mobile deposit |
| US9129340B1 (en) | 2010-06-08 | 2015-09-08 | United Services Automobile Association (Usaa) | Apparatuses, methods and systems for remote deposit capture with enhanced image detection |
| US8488886B2 (en) * | 2010-08-02 | 2013-07-16 | Celartem, Inc. | Font matching |
| AU2012202173B2 (en) | 2011-04-18 | 2013-09-05 | Castle Bookkeeping Wizard Pty Ltd | System and method for processing a transaction document including one or more financial transaction entries |
| US10380565B1 (en) | 2012-01-05 | 2019-08-13 | United Services Automobile Association (Usaa) | System and method for storefront bank deposits |
| US10552810B1 (en) | 2012-12-19 | 2020-02-04 | United Services Automobile Association (Usaa) | System and method for remote deposit of financial instruments |
| CN105612534B (en) * | 2013-05-24 | 2019-07-12 | 生命技术公司 | Method and system for the consumer goods in detection system |
| US11138578B1 (en) | 2013-09-09 | 2021-10-05 | United Services Automobile Association (Usaa) | Systems and methods for remote deposit of currency |
| US9286514B1 (en) | 2013-10-17 | 2016-03-15 | United Services Automobile Association (Usaa) | Character count determination for a digital image |
| US11295308B1 (en) | 2014-10-29 | 2022-04-05 | The Clearing House Payments Company, L.L.C. | Secure payment processing |
| US10402790B1 (en) | 2015-05-28 | 2019-09-03 | United Services Automobile Association (Usaa) | Composing a focused document image from multiple image captures or portions of multiple image captures |
| US11042882B2 (en) | 2015-07-01 | 2021-06-22 | The Clearing House Payments Company, L.L.C. | Real-time payment system, method, apparatus, and computer program |
| US11694168B2 (en) | 2015-07-01 | 2023-07-04 | The Clearing House Payments Company L.L.C. | Real-time payment system, method, apparatus, and computer program |
| US10346702B2 (en) | 2017-07-24 | 2019-07-09 | Bank Of America Corporation | Image data capture and conversion |
| US10192127B1 (en) | 2017-07-24 | 2019-01-29 | Bank Of America Corporation | System for dynamic optical character recognition tuning |
| US11030752B1 (en) | 2018-04-27 | 2021-06-08 | United Services Automobile Association (Usaa) | System, computing device, and method for document detection |
| US11436577B2 (en) | 2018-05-03 | 2022-09-06 | The Clearing House Payments Company L.L.C. | Bill pay service with federated directory model support |
| US11900755B1 (en) | 2020-11-30 | 2024-02-13 | United Services Automobile Association (Usaa) | System, computing device, and method for document detection and deposit processing |
Family Cites Families (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| SE422714B (en) * | 1979-01-16 | 1982-03-22 | Ibm Svenska Ab | DEVICE WITH OPTICAL SIGN SOLUTION COMPUTER, WHICH CONTROL COMPUTER CALCULATES STARTING POINTS FOR IDENTIFYING A SIGN |
| US4308523A (en) * | 1980-02-04 | 1981-12-29 | Compuscan, Incorporated | Apparatus and method for character recognition |
| NL183790C (en) * | 1980-11-21 | 1989-01-16 | Nederlanden Staat | METHOD FOR CHARACTER SEGMENTATION. |
| GB2103449B (en) * | 1981-06-29 | 1985-05-30 | Nippon Telegraph & Telephone | Method and apparatus for gray level signal processing |
| US4783829A (en) * | 1983-02-23 | 1988-11-08 | Hitachi, Ltd. | Pattern recognition apparatus |
| JPS59226981A (en) * | 1983-06-08 | 1984-12-20 | Fujitsu Ltd | Method and device for pattern matching |
| US4471046A (en) * | 1983-08-04 | 1984-09-11 | Eastman Kodak Company | 4-Hydroxyalkyl-substituted 3-pyrazolidinone electron transfer agents |
| US4625330A (en) * | 1983-12-19 | 1986-11-25 | Ncr Corporation | Video enhancement system |
| ATE56549T1 (en) * | 1984-04-10 | 1990-09-15 | British Telecomm | PATTERN RECOGNITION DEVICE. |
| US4742556A (en) * | 1985-09-16 | 1988-05-03 | Davis Jr Ray E | Character recognition method |
| US5014327A (en) * | 1987-06-15 | 1991-05-07 | Digital Equipment Corporation | Parallel associative memory having improved selection and decision mechanisms for recognizing and sorting relevant patterns |
| US4856075A (en) * | 1987-10-05 | 1989-08-08 | Eastman Kodak Company | Image discrimination |
| US4850026A (en) * | 1987-10-13 | 1989-07-18 | Telecommunications Laboratories Dir. Gen'l Of Telecom. Ministry Of Communications | Chinese multifont recognition system based on accumulable stroke features |
| US4949392A (en) * | 1988-05-20 | 1990-08-14 | Eastman Kodak Company | Document recognition and automatic indexing for optical character recognition |
-
1990
- 1990-12-28 US US07/635,899 patent/US5091968A/en not_active Expired - Lifetime
-
1991
- 1991-08-23 CA CA002049758A patent/CA2049758C/en not_active Expired - Fee Related
- 1991-12-19 EP EP91311790A patent/EP0492979B1/en not_active Expired - Lifetime
- 1991-12-19 DE DE69131006T patent/DE69131006T2/en not_active Expired - Fee Related
- 1991-12-26 JP JP3356742A patent/JPH04361391A/en active Pending
Also Published As
| Publication number | Publication date |
|---|---|
| JPH04361391A (en) | 1992-12-14 |
| EP0492979B1 (en) | 1999-03-17 |
| CA2049758A1 (en) | 1992-06-29 |
| EP0492979A3 (en) | 1994-02-23 |
| EP0492979A2 (en) | 1992-07-01 |
| DE69131006T2 (en) | 1999-10-07 |
| US5091968A (en) | 1992-02-25 |
| DE69131006D1 (en) | 1999-04-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CA2049758C (en) | Optical character recognition system and method | |
| US5367578A (en) | System and method for optical recognition of bar-coded characters using template matching | |
| EP0063454B1 (en) | Method for recognizing machine encoded characters | |
| EP0543592B1 (en) | Optical word recognition by examination of word shape | |
| US5687253A (en) | Method for comparing word shapes | |
| US6335986B1 (en) | Pattern recognizing apparatus and method | |
| US5640466A (en) | Method of deriving wordshapes for subsequent comparison | |
| US4630308A (en) | Character reader | |
| US5197107A (en) | Character recognition apparatus | |
| US5081690A (en) | Row-by-row segmentation and thresholding for optical character recognition | |
| US5272766A (en) | OCR system for recognizing user-specified custom fonts in addition to standard fonts using three-layer templates | |
| Lehal et al. | Feature extraction and classification for OCR of Gurmukhi script | |
| US5926564A (en) | Character recognition method and apparatus based on 0-1 pattern representation of histogram of character image | |
| US5563958A (en) | System and method for optical character recognition bar-coded characters | |
| JP2008250754A (en) | Character string recognition method and device | |
| JP5630689B2 (en) | Character recognition method and character recognition device | |
| US5077809A (en) | Optical character recognition | |
| US8200016B2 (en) | Method and apparatus for character string recognition | |
| EP0495607B1 (en) | A method and apparatus for representing an unrecognised character in optical character recognition means for subsequent recognition | |
| Ullmann | Picture analysis in character recognition | |
| JP2917427B2 (en) | Drawing reader | |
| JPH11328309A (en) | Method and device for optical character read | |
| JP3710164B2 (en) | Image processing apparatus and method | |
| JPH0877293A (en) | Character recognition device and generating method for dictionary for character recognition | |
| JPS61269780A (en) | Character recognizing system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| EEER | Examination request | ||
| MKLA | Lapsed |