WO1995012280A1

WO1995012280A1 - Transmission system and a method of coding and decoding a bidimensional signal

Info

Publication number: WO1995012280A1
Application number: PCT/FI1993/000445
Authority: WO
Inventors: Timo Kilpinen
Original assignee: Telecom Finland Oy
Priority date: 1993-10-29
Filing date: 1993-10-29
Publication date: 1995-05-04

Abstract

The object of the invention is a method for e.g. a point-to-point image transfer. In the method there are means at both ends to enable the image transfer whereby the transfer takes place as digital along transmission routes. The image data stored in the frame storage is converted into points in the transformed range in the processor unit for the transfer of the image data.

Description

TRANSMISSION SYSTEM AND A METHOD OF CODING AND DECODING A BIDIMENSIONAL SIGNAL

The invention relates to transmission system/method and a method of coding and decoding a bidimensional signal.

In a digital video compression techniques matrices of pixels in an appropriate signal format are converted into other rep¬ resentation selective sampling of the chrominance components. The signals are then subjected to a discrete cosine transform. The picture should remain in the quality detectable by the hu¬ man eye. In order to achieve high compression ratios and still maintain a high image quality computationally intensive algo¬ rithms must be relied on. Algorithms should be run in real time. In the applications the information which must be stored to reproduce a video picture is still quite enormous. A plura¬ lity of video images is to be generated in sequence to repli¬ cate either changes in images or data.

The image or data transfer method operates e.g. in a point-to- point transmission. The receiver and the transmitter can have different or substantially similar^"means, as presented in pic¬ ture 1, depending on whether the intention is to send or re¬ ceive, comprising a common PC/AT-computer 1 equipped with a VGA-display 2 and video- and data transmissions cards (e.g. X21-interface card) as well as a colour video camera 3 and a video monitor 4. The digitized still images are fed into the system as a video or RGB-signal according to the PAL-standard. Such picture sources are e.g. a video camera, a video recorder and a still-video camera. A special X-ray scanner is used to read the x-ray pictures and the pictures can be watched on the computer display. If satisfied with VGA-pictures, these can be watched on the computer display. The data transmission can take place e.g. at the speed of the digital telephone net, e.g. at the speed of 64 kbit/s. The image is compressed into the transmission channel. As transmission routes can be used video conference links, fixed data links as well as subscriber links of the digital tele-net DIGINET and ISDB or even modem links .

The resolution of the VGA-image transmitted to the computer display can be e.g. 320x200 points, so that in the display there can simultaneously be 256 different colour shades out of 32000 colour shades. The resolution of the still image trans¬ mitted to the video monitor display can be e.g. 768x576 points, of which each can be displayed in 32000 colour shades. Due to the big storage space requirement of the still image the efficient compression techniques according to the JPEG standard is used in the transmission, in which the storage space required by the image is packed into 1/40-1/20 parts of the original without any visually detectable deterioration of the image quality, as Gregory K. Wallace presented in his pa¬ per Overview of the JPEG (=Joint Photographic Experts Group) (ISO/CCIT) Still image Compression Standard "For presentation and distribution at Visual communications and Image Processing '89, SPIE, Philadelphia, November 1989. In the JPEG the pictu¬ re is unpacked into chrominance and colour saturation com¬ ponents, which are processed separately. From the image infor¬ mation is removed the part that is anyhow undetectable to the human eye. The data is coded into a smaller space. The total transmission time of the image, including the compression ti¬ me, decreases to between 15-100 s depending on the effect and the required storage space of the computer used, e.g. from 864 kbytes to 20-70 kbytes. The images can be optionally stored in the original, compressed or VGA-form. The images thus stored can be watched and sent later along the data links. The images can also be modified, cut, combined and/or subtitled with an image processing program. Because of the video image, a great amount of data has to be stored in the storage to enable a good image quality. When the amount of data to be stored can be decreased without affecting the picture, the image trans¬ mission is expedited.

It is common practice with the coding and decoding to divide a bidimensional signal into blocks each having a predetermined block size and then code and decode it on a block basis. The problem of this kind of coding and decoding scheme is that discontinuous changes in the signal so called block boundary artifacts occur at the boundaries of adjoining blocks. For that it is possible to use overlapping regions e.g. dividing an image signal into blocks each having 2N x 2N pixels N being a natural number and overlapping adjacent blocks by half the block size in four directions which are perpendicular to the four sides thereof. An input pixel signal representative of 2Nx2N points is coded into a transformed signal representative of NxN points by transform equations and then the transformed signal is transmitted. An inverse procedure is used. Subse¬ quently, this inversed transformed signal and signals of tr e associated regions of the overlapping blocks are added to pro¬ duce a decoded signal. On the other hand it is known that when a bidimensional picture signal is divided into blocks and then coded and decoded, increasing the block size in regions where the change is insignificant and decreasing it in region where the change is significant is successful in enhancing efficient coding and decoding. It is presented that a procedure which splits a picture signal into non-overlapping bidimensional blocks of variable sizes and effects transform and coding in each block of the blocks. Further above coding and decoding scheme suffers block boundary artifacts since they divide an input signal rn→ blocks such that the blocks do not overlap one another.

EP-A-0467054 describes a method to provide a coding and deco¬ ding method and apparatus of the type splitting a bidimen¬ sional picture signal into picture blocks and coding and deco¬ ding on a block basis and capable of decoding an input signal correctly even when blocks overlap and have variable block size. It describes coding and decoding a bidimensional signal comprising the steps of at a coding side, dividing a bidimen¬ sional input signal into rectangular blocks of plurality of sizes, providing window functions on a direction and block size basis, calculating, by dividing two-dimensional signal different size block by a discrete cosine-transformation. The block of interest has an MxN size, a transformed signal from input signals by means of a particular transform equation. The block division information and the transformed signals are transmitted to a decoding side. The window functions cause a specific stress to the signal. The publication aims at solving the image transfer by calculating also overlapping image blocks. The invention aims at providing a solution of how to minimize the load in the processor storage and how to expedite the transmission of the pixels.

The problem of the above mentioned image transfers is, despite the JPEG-standard, the slowness. The reason for this is that numerous calculation operations, especially storage capacity loading multiplications, have to be done, when the data is transmitted between the frame storage and the processor unit and is compressed. The data of each pixel point is separately retrieved from the frame storage into the processor to be transformed into transformed spatial points. This stage has been improved by us. To solve this the invention is characte¬ rized by what is disclosed in the novelty part of claim 1.

The invention relates to improving the speed of the image transmission, so that it takes place substantially in real time and without substantially deteriorating the image quali¬ ty. A PC/AT-class computer can be used.

The invention is described with reference to the enclosed dra¬ wings, in which

fig. 1 presents an application example of the image transfer method's transmission links of the subject invention,

fig. 2 presents the transmission of the pixel into the process unit by visualizing the data of one pixel in the frame storage and

fig. 3a, b, c and d present each the conversion of the image spatial data of a selected group into transformed spatial points. A standard video signal is first brought e.g. from the camera via the equipment entry to a analog/digital (A/D-) converter, where the image signal is digitized. The digitized image in¬ formation is read into the frame storage i.e. to the image digitizing card 11. The display visualizes the contents of the image digitizing card. The computer processor 10 controls the operations including a sufficient storage and the programming intelligence. In accordance with the invention the data con¬ tained in the frame storage is read as per pixel, e.g. in the form of a 8x8 matrice into the processor, where the digitized data is compressed. In the method a pixel is selected, all the pixel point information is fed into the process unit, the pi¬ xel is divided into groupr i, q, r and s, which ups by se¬ parate decentralized proct .-sing are converted ii points in the transformed range P, Q, R and S, which thereafter are quantized and coded in coder 12 to be transmitted along the transmission link into the required terminal by coding the transformed spatial points in the coder 12 and by transmitting into the transmitter buffer and via the output into the trans¬ mission route 13, via which the image information is forwarded to the receiver. In the quantiization the aim is to remove information undetectable to the human eye.

As can be seen from fig. 2, each spatial pixel is fetched from the frame storage 11 simultaneously into the processor unit 10. In the following is presented how a two dimensional cosine transform is made to the 8x8 block.

Discrete cosine transform

The NxN two dimensional DCT is defined as:

2 N-l N-l (2x+l)m*π (2y+l)n*π

F ^β C C Σ Σ f cos[ ] cos[ ] m,n N m n x=0 y=0 x,y 2N 2N

(1) with m,n,x,y = 0,1,2....,N-1 where x,y are spatial coordinates in the pixel domain m,n are coordinates in the transform domain 1/^2^" for m,n = 0

C , C m n 1 otherwise The inverse DCT (IDCT) is defined as:

2 N-l N-l (2x+l)m*π (2y+l)n*π

: = ∑ ∑ C C F cos[ ] cos[ ] x,y N m=0 n=0 n m,n 2N 2N

(2)

The two dimensional DCT can also be expressed in matrix formulation. When N = 8 we get for 8x8 FDCT:

1 T

F = - A * f * A (3) 4 and for 8x8 IDCT

1 T f = - A * F * A (4) 4 where c4 c4 c4 c4 c4 c4 c4 c4 cl c3 c5 c7 -c7 -c5 -c3 -cl c2 c6 -c6 -c2 -c2 c6 c6 c2

A = c3 -c7 -cl -c5 c5 cl c7 -c3 c4 -c4 -c4 c4 c4 -c4 -c4 c4 c5 -cl c7 c3 -c3 -c7 cl -c5 c6 -c2 c2 -c6 -c6 c2 -c2 c6 c7 -c5 c3 -cl cl -c3 c5 -c7

c = cos(i*π/16) ; A*A = 41 i

E is a trivial transformation matrix E * E = I m,n m,n m,n when operating to the right it changes rows m ans n when operating to the left it changes columns m and n

Reordering the matrix A gives:

A4 A4

E35*E24*E56*E12*A*E47*E56 = (5)

B4 -B4

T

A4 B4

E56*E47*A *E12*E56*E24*E35 = T

A4 -B4 (6) where shared matrices are c4 c4 c4 c4 cl c3 c5 c7 T

A4 = c2 c6 -c6 -c2 B4 = c3 -c7 -cl -c5 = B4 c4 -c4 -c4 c4 c5 -cl c7 c3 c6 -c2 c2 -c6 c7 -c5 c3 -cl (7)

A4 * A4 = 2 1 B4 * B4 = 2 I

2cl 0 0 0 1 c2 c4 c6 1 0 0 0

L4*B4 = 0 2c3 0 0 *B4 = 1 c6 -c4 -c2 1 1 0 0

0 0 2c5 0 1 -c6 -c4 c2 0 1 1 0

0 0 0 2c7 1 -c2 c4 -c6 0 0 1 1

From (3) we get

E35*E24*E56*El_** *E12*E56*E24*E35 = (1/4)*E35*E2 ≤56*E12*A*E47*E56*E47*f*

E47*E56*E56*E47*A *E12*E56*E24*E35

T

A4 A4 A4 B4

(1/4) *E56*E47*f*E47*E56* T (8)

B4 -B4 A4 -B4

From (4) we get

E56*E47*f*E47*E56 =

T (1/4)*E56*E47*A *E12*E56*E24*E35*E35*E24*E56*E12*F*

E12*E56*E24*E35*E35*E24*E56*E12*A*E47*E56 =

T A4 B4 A4 A4

(1/4) T *E35*E24*E56*E12*F*E12*E56*E24*E35* A4 -B4 B4 -B4 (9)

Reordering the block in pixel domain gives: fOO fOl f02 f03 f07 f06 f05 f04 flO fll fl2 fl3 fl7 fl6 fl5 fl4 f20 f21 f22 f23 f27 f26 f25 f24 f30 f31 f32 f33 f37 f36 f35 f34 f70 f71 f72 f73 f77 f76 f75 f74 f60 f61 f62 f63 f67 f66 f65 f64 f50 f51 f52 f53 f57 f56 f55 f54

f40 f41 f42 f43 f47 f46 f45 f44

p q

(10) r s

i,j = 0,1,2,3 P(i,j⁾ f(i,j) q(i#j) f(i,7-j) r(i,j) f(7-i,j) s(i,j) f(7-i,7-j)

Reordering the block in transform domain gives:

E35*E24*E56*E12*F*E12*E56*E24*E35 =

FOO F02 F04 F06 FOl F03 F05 F07 F20 F22 F24 F26 F21 F23 F25 F27 F40 F42 F44 F46 F41 F43 F45 F47 F60 F62 F64 F66 F61 F63 F65 F67 P Q

(ID

FlO F12 F14 F16 Fll F13 F15 F17 R S F30 F32 F34 F36 F31 F33 F35 F37 F50 F52 F54 F56 F51 F53 F55 F57 F70 F72 F74 F76 F71 F73 F75 F77

i,j = 0,1,2,3

When (10) and (11) are cast in equations (8) and (9) we get

T

P Q A4 A4 P q A4 B4

FDCT: T (12) R S B4 -B4 r s A4 -B4

T p q A4 B4 P Q A4 A4

IDCT: 4 T (13) r s A4 -B4 R S B4 -B4 After multiplication of shared matrices in (12) and (13) we get

P Q A4(p+q+r+s)A4 A4(p-q+r-s)B4

FDCT: 4 T (14) R S B4(p+q-r-s)A4 B4(p-q-r+s)B4

p q

IDCT: 4 r s

T T T T

A4 PA4+A4 QB4+B4RA4+B4SB4 A4 PA4-A4 QB4+B4RA4-B4SB4

T T T T (15)

A4 PA4+A4 QB4-B4RA4-B4SB4 A4 PA4-A4 QB4-B4RB4+B4SB4

FDCT:

4P = A4(p+q+r+s)A4

4Q = A4(p-q+r-s)B4

T

4R = B4(p+q-r-s)A4

4S = B4(p-q-r+s)B4 (16)

IDCT:

T T 4p = A4 PA4 + A4 QB4 + B4RA4 + B4SB4

T T 4q = A4 PA4 - A4 QB4 + B4RA4 - B4SB4

T T 4r = A4 PA4 + A4 QB4 - B4RA4 - B4SB4

T T 4s = A4 PA4 - A4 QB4 - B4RA4 + B4SB4 (17) Walsh-Hadamard:

A 1 1 1 1 a a

B 1 -1 1 -1 b b

* = H4 *

C 1 1 -1 -1 c c

D 1 -1 -1 1 d d

A B 1 1 a b 1 1 a b

= H2 H2

C D 1 -1 c d 1 -1 c d

* H4 = 4 I H2 * H2 - 2 I

FDCT:

P P 4P = A4 p A4 q = H4 * q r r 4Q = A4 q B4 s s

4R = B4 r A4

4S = B4 s B4 (18)

IDCT:

P A4 P A4

T q A4 Q B4

= H4 * r B4 R A4

B4 S B4 (19)

Now both processes have been dealt in four parts which are not dependent on each other and the multiplications shall be calculated only between 4x4 matrices.

Calculation of 4P = A4 p A4 A2 A2 T A2 B2

E12*A4*E23 = E23*A4 *E12 = B2 -B2 A2 -B2 c2 c6 where A2 = c4*H2 ; B2 = c6 -c2

1 0 1 0

B2 = A2*B2* *B2*A2 0 -1 0 -1

T

4*E12*P*E12 = El2*A4*E23*E23*p*E23*E23*A4 *E12

POO P02 POl P03 pOO pOl p03 p02 P20 P22 P21 P23 A2 A2 plO pll pl3 pl2 A2 B2

PlO P12 Pll P13 B2 -B2 JO p31 p33 p32 A2 -B2 P30 P32 P31 P33 p20 p21 p23 p22

After the Hadamard transform pOO pOl plO pll pOO pOl plO pll p03 p02 pl3 pl2 = H4 p03 p02 pl3 pl2 p30 p31 p20 p21 p30 p31 p20 p21 p33 p32 p23 p22 p33 p32 p23 p22 we get

POO P02 pOO pOl pOO pOl

= 2A2 A2 = H2 H2 P20 P22 plO pll plO pll

POl P03 p03 p02 p03 p02 1 0

=2A2 B2=H2 H2*B2 P21 P23 pl3 pl2 pl3 pl2 0 -1

PlO P12 p30 p31 1 0 p30 p31

=2B2 A2= B2*H2 H2 P30 P32 p20 p21 0 -1 p20 p21

Pll P13 p33 p32

= 2B2 B2 P31 P33 p23 p22 Pll -P13 1 c4 p33+p22 p32-p23 P33 P31 1 -c4 p33+p32+p23-p22 -p33+p32+p23+p22

Block diagrams and their explanations for operators described in Figs. 3a to 3d:

* 1 0

B = *B2

0 -1 Calculation of 4Q = A4 q B4

QOO Q20 -Q30 -Q13

-QOl -Q21 Qll Q32 = B4*G

-Q02 -Q22 -Q12 Q31

Q03 Q23 Q33 -QlO where gOO = = q01+qll+q21+q31+q02+ql2 -2+q32 glO = = q00+ql0+q20+q30+q03+ql3+q23+q33 g20 = = q00+ql0+q20+q30-q03-ql3-q23-q33 g30 = = q01+qll+q21+q31-q02-ql2-q22-q32 gOl = q0l-qll-q2l+q2¹.+q02-ql2-q22+q32 gll = q00-ql0-q20+q30+q03-ql3-q23+q33 g21 = q00-ql0-q20+q30-q03+ql3+q23-q33 g31 = q01-qll-q21+q31-q02+ql2+q22-q32 g02 = ql0-q20+qll-q21-q02+q32-q03+q33 gl2 = ql0-q20-q01+q31+ql2-q22+q03-q33 g22 = -q00+q30+qll-q21+q02-q32+σl3-q23 g32 = -q00+q30+q01-g31+ql2-q22- 13+q23 g03 = -qiϋ+q20+qll~ l-q02+q32+q03-q33 gl3 = ql0-q20+q01-q31-ql2+q22+q03-q33 g23 = -q00+q30-qll+q21-q02+q32+ql3-q23 g33 = q00-q30+q01-q31+ql2-q22+ql3-q23

After transformation qOO q03 qOl q02 qOO q03 qOl q02 ql3 qlO ql2 qll = H4* ql3 qlO ql2 qll q23 q20 q22 q21 q23 q20 q22 q21 q30 q33 q31 q32 q30 q33 q31 q32

q01+q02 q31+q32 -qlO-qll σ^~ -qll

G = q00+q03 q30+q33 q20-ql2 c ql2 q30-q33 q00-q03 q21-ql3 -c gl3 q31-q32 q01-q02 q22-q23 q- . 23

After transformation qOO q30 qOl q31 qlO q20 q21 q22 qOO q30 qOl q31 qlO q20 q21 q22

=H2 q03 q33 q02 q32 qll ql2 ql3 q23 q03 q33 q02 q32 qll ql2 ql3 q23

qOl q31 -qlO qll

G = qOO q30 ql2 q20 q33 q03 ql3 -q21 q32 q02 q23 q22 QOO Q20 -Q30 -Q13 1 c2 c4 c6 1 0 0 0 -QOl -Q21 Qll Q32 1 c6 -c4 -c2 1 1 0 0 *G -Q02 -Q22 -Q12 Q31 1 -c6 -c4 c2 0 1 1 0

Q03 Q23 Q33 -QlO

1 -c2 c4 -c6 0 0 1 1

QOO Q20 -Q30 -Q13 1 c4 c2 c6 1 0 0 0 3*L4* -QOl -Q21 Qll Q32 1 -c4 c6 -c2 0 1 1 0 *G -Q02 -Q22 -Q12 Q31 1 c4 -c2 -c6 1 1 0 0

Q03 Q23 Q33 -QlO

1 -c4 -c6 c2 0 0 1 1

1 c4 gOO gOl g02 g03

B2 1 -c4 gl0+g20 gll+g21 gl2+g22 gl3+g23

1 c4 gOO+glO gOl+gll g02+gl2 g03+gl3

-B2 1 -c4 g20+g30 g21+g31 g22+g32 g23+g33

qOl q31 -qlO qll

K B2 q00+q33 q30+q03 ql2+ql3 q20-q21 qOl+qOO q31+q30 ^■ql0+ql2 qll+q20

K -B2 q33+q32 q03+q02 ql3+q23 -q21+q22

L c 1 1 0 where = H2 *

1 -c4 0 c4

cl 0 QOO Q20 qOl q31 qOl+qOO q31+q30

=K* +B2*

0 2c3 -QOl -Q21 q00+q33 q30+q03 q33+q32 q03+q02

cl 0 -Q30 -Q13 -qlO qii -ql0+ql2 qll+q20

=K* +B2*

0 2c3 Qll Q32 ql2+ql3 q20-q21 ql3+q23 -q21+q22

c7 0 Q03 Q23 qOl q31 qOl+qOO q31+q30

=K* -B2*

0 2c5 -Q02 -Q22 q00+q33 q30+q03 q33+q32 q03+q02

c7 0 Q33 -QlO -qlO qll -ql0+ql2 qll+q20

=K* -B2*

0 2c5 -Q12 Q3] q^] .2+ql3 q2C )-q21 ql3+q23 -q21+q22 T

Calculation of 4R = B4 r A4

R00 R02 -R03 -R31

-RIO -R12 Rll R23 = B4*G

-R20 -R22 -R21 R13

R30 R32 R33 -R01 where gOO = rl0+rll+rl2+rl3+r20+r21+r22+r23 glO = r00+r01+r02+r03+r30+r31+r32+r33 g20 = r00+r01+r02+r03-r30-r31-r32-r33 g30 = rl0+rll+rl2+rl3-r20-r21-r22-r23 gOl = rl0-rll-rl2+rl3+r20-r21-r22+r23 gll = r00-r01-r02+r03+r30-r31-r32+r33 g21 = r00-r01-r02+r03-r30+r31+r32-r33 g31 = rl0-rll-rl2+rl3-r20+r21+r22-r23 g02 = r01-r02+rll-rl2-r20+r23-r30+r33 gl2 = r01-r02-rl0+rl3+r21-r22+r30-r33 g22 = -r00+r03+rll-rl2+r20-r23+r31-r32 g32 = -r00+r03+rl0-rl3+r21-r22-r31+r32 g03 = -r01+r02+rll-rl2-r20+r23+r30-r33 gl3 = r01-r02+rl0-rl3-r21+r22+r30-r33 g23 = -r00+r03-rll+rl2-r20+r23+r31-r32 g33 = r00-r03+rl0-rl3+r21-r22+r31-r32

After transformation rOO r30 rlO r20 rOO r30 rlO r20 r31 rOl r21 rll H4* r31 rOl r21 rll r32 r02 r22 rl2 r32 r02 r22 rl2 r03 r33 rl3 r23 r03 r33 rl3 r23

rl0+r20 rl3+r23 -rOl-rll rOl-rll

G - r00+r30 r03+r33 r02-r21 r02+r21 r03-r33 r00-r30 rl2-r31 -rl2-r31 rl3-r23 rl0-r20 r22-r32 r22+r32

After transformation O r03 rlO rl3 rOl r02 rl2 r22 rOO r03 rlO rl3 rOl r02 rl2 r22

=H2 0 r33 r20 r23 rll r21 r31 r32 r30 r33 r20 r23 rll r21 r31 r32

rlO rl3 -rOl rll

G = rOO r03 r21 r02 r33 r30 r31 -rl2 r23 r20 r32 r22 ROO R02 -R03 -R31 -1 1 c2 c4 c6 1 0 0 0 -RIO -R12 Rll R23 =L4 * 1 c6 -c4 -c2 1 1 0 0 *G -R20 -R22 -R21 R13 1 -c6 -c4 c2 0 1 1 0

R30 R32 R33 -ROl 1 -c2 c4 -c6 0 0 1 1

ROO R02 -R03 -R31 1 c4 c2 c6 1 0 0 0 3*L4* -RIO -R12 Rll R23 1 -c4 c6 -c2 0 1 1 0 *G -R20 -R22 -R21 R13 1 c4 -c2 -c6 1 1 0 0

R30 R32 R33 -ROl

1 -c4 -c6 c2 0 0 1 1

23 l3 33

l2 02

22

cl 0 ROO R02 rlO rl3 rlO+rOO rl3+r03

=K* +B2* 0 2c3 ^■RIO -R12 r00+r33 r03+r30 r33+r23 r30+r20

cl 0 -R03 -R31 -rOl rll -r01+r21 rll+r02

=K* +B2* 0 2c3 Rll R23 r21+r31 r02-rl2 r31+r32 -rl2+r22

c7 0 R30 R32 rlO rl3 rlO+rOO rl3+r03

=K* -B2* 0 2c5 -R20 -R22 r00+r33 r03+r30 r33+r23 r30+r20

c7 0 R33 -ROl -rOl rll -r01+r21 rll+r02

=K* -B2* 0 2c5 R21 R13 r21+r31 r02-rl2 r31+r32 -rl2+r22

Calculation of 4S = B4 s B4 S00 -S30 S20 S31 1 c2 c4 c6

Sll S21 -S01 S23 1 c6 -c4 -c2 * G

S22 -S12 S32 -S10 1 -c6 -c4 c2

S33 S03 S13 S02 1 -c2 c4 -c6 where gOO = s00+sll+s22+s33 glO = s00+sl0+s01+s21+sl2+s32+s23-s33 g20 = Sl0+s20+s01+s31+s02-s32+sl3-s23 g30 = s20+s30+sll-s31+s02-s22+s03-sl3 gOl = s30-s21+sl2-s03 gll = -s20+s30+sll+s31-s02-s22+s03+sl3 g21 = Sl0-s20-s01+s31+s02+s32-sl3-s23 g31 = -s00+sl0+s01-s21-sl2+s32+s23+s33 g02 = -sl0+s31+s02+s23 gl2 = -Sl0+s30+s01-sll+s22+s32+s03-s23 g22 = s00+s30-sll+s21-sl2-s22-s03+s33 g32 = s00+s20-s21+s31-s02-sl2-sl3-s33 g03 = 20-s01+s32+sl3 gl3 = - 10-s30+s01+sll-s22+s32-s03-s23 g23 = s00-s30-sll-s21+sl2-s22+s03+s33 g33 = -s00+s20+s21+s31-s02+sl2-sl3+s33

After transformations: sOO sll sl2 i 13 s02 slO s20 sOl sOO sll sl2 s03 s02 slO s20 sOl

=H2 s33 s22 s21 s30 s31 s23 sl3 s32 s33 s22 s21 s30 s31 s23 sl3 s32

sOO s33 s22 s21 s02 sl3 sOl s20 sOO s33 s22 s21 s02 sl3 sOl s20

=H2 sll sl2 s03 s30 s23 s31 slO s32 sll sl2 s03 s30 s23 s31 slO s32

sll s02 s22 slO s31 sOl sll s02 s22 slO s31 sOl

=H2 s21 s20 sl3 s03 sl2 s33 s21 s20 sl3 s03 sl2 s33

We get for matrix G: sOO s30 s23 s32

G = sOl sl3 s03 slO s02 s20 s21 sll s22 s33 s31 sl2 gOO gOl g02 g03

SOO -S30 S20 S31 K B2 Sll S21 -SOl S23 g20 g21 g22 g23

8 E23

S22 -S12 S32 -SlO glO gll gl2 gl3 S33 S03 S13 S02 K -B2 g30 g31 g32 g33 sOO s30 s23 s32

SOO -S30 S20 S31 K B2 Sll S21 -SOl S23 s02 s20 s21 sll

S33 S03 S13 S02 sOl sl3 s03 slO S22 -S12 S32 -SlO K -B2 s22 s33 s31 sl2

SOO -S30 sOO s30 sOl sl3

= K + B2 * Sll S21 s02 s20 s22 s33

S20 S31 s23 s32 s03 slO

= K * + B2 * -SOl S23 s21 sll s31 sl2

S33 S03 sOO s30 sOl sl3

= K * - B2 * S22 -S12 s02 s20 s22 s33

S13 S02 s23 s32 s03 slO

= K * - B2 * S32 -SlO s21 sll s31 sl2

The mark f presents the luminance value of the image. A descri¬ bes the coefficient matrice of the cosine transform and E the elementary transformation operator, by which the order of the matrice lines or columns can be changed. The fig. 3a, b, c and d present schematically the conversion of p, q, r and s into points in the transformed range, P, Q, R and S, correspondingly. Fig. 3a, which depicts the conversion of p, is presented on one page. The fig. 3b, c and d, which depict the conversion of the image points q, r and s, are presented on two pages. The reverse conversion is made correspondingly.

Experiments have been made on the 64 kbit/s-channel. When the amount of the image points is e.g. 720 x 578 and the colours are three and the PC clock frequency is 20 MHz, the transmission of a 1,25 Mbytes image from the transmitter to the receiver takes with the old method, i.e. point-to-point calculation, about 190 s, when again, with the method according to the invention, the time is less than 12 s. A considerable advantage is that with the subject method a 16-time improvement of the speed is ac¬ hieved, although the image quality is still according to the JPEG-standard.

As one application of the invention can be mentioned i.a. fully- digital hdtv-televisions, in which image information is sent from one point to several points.

The invention has been described above by reference to one of its favorable examples of application. This is not to be consi¬ dered as so limited, but the invention can be modified within the scope of protection defined by the enclosed patent claims.

Claims

1. A method for image transfer, in which method the receiver and the transmitter have means to enable the data transmission, the transmission taking place as digital along transmission routes, the image data stored in the frame storage (11) is converted into points in the transformed range in the processor unit (10) for the transmission of the image information, c h a r a c ¬ t e r i z e d in that selecting a pixel and changing the pixel points into a predeter¬ mined order; feeding all data of the pixel points into the processor; dividing the pixel into equally big groups (p, q, r, s); calculating for the pixel groups (p, q, r, s) separately points in the transformed range (P, Q, R, S); quantizing, coding and transmitting these to the transmission route.

2. A method according to claim 1, c h a r a c t e r i z e d in that converting the groups (p, q, r, s) substantially simulta¬ neously into points in the transformed range.

3. A method according to claim 1 or 2, c h a r a c t e ¬ r i z e d in that decentralizing groups (p, q, r, s) into sub¬ groups.

4. A method according to the claims 1, 2 or 3, c h a r a c ¬ t e r i z e d in that rearranging the pixel into a predetermined order.

5. A method according to claim 4, c h a r a c t e r i z e d in that presenting the pixel groups (p, q, r, s) in a matrice form.

6. A method according to one of the claims above, c h a r a c ¬ t e r i z e d in that it is used in fully digital hdtv-systems.

7. Method to code and decode a bidimensional signal using the discrete cosine transformation, c h a r a c t e r i z e d in that digitizing the information and sending the digitized infor¬ mation into the processor, dividing a bidimensional input signal into 8x8 blocks of same size in horizontal and vertical direc¬ tions, arranging the block again in four equal size 4x4 blocks, chan¬ ging the block by distributed processing with Hadamard-trans¬ forms, so that these 4x4 matrices are independent of each other, and these triple matrice products A4pA4T, A4qB4, B4A4T, B4sB4 decay further to two-matrice products so that the total amount of scalar multiplications is 94.