CN102663422B - Floor layer classification method based on color characteristic - Google Patents

Abstract

The invention discloses a floor layer classification method based on a color characteristic. In a HSV color space, extraction of a floor color moment characteristic is performed on tone and saturation components, a high weight is given to tone characteristic data and an influence of a texture characteristic on the classification is reduced. A brightness component is removed. The influence of an industrial on-site illumination condition on the floor is avoided. When carrying out on-line classification on a unknown floor sample, determination is performed on the unknown floor sample from coarse to fine layer by layer through using a offline multilayer classification basis database which is established in advance and adopting a K nearest adjacent-shortest distance decision. The method of the invention has the following advantages that: a floor manufacturer can accurately classify floor classes; floor production efficiency can be increased and artificial cost can be reduced.

Description

The floor hierarchy classification method of color-based feature

Technical field

The present invention relates to a kind of floor hierarchy classification method of color-based feature, be applied to the floor industry spot classification of new production floor sample is judged.

Background technology

Along with the fast development of economy, the demand on floor is also more and more, and the floor is categorized into for a problem in the urgent need to address fast and accurately in production run.At present, the classification on floor realizes the main artificial visually examine of leaning on, and the impact of its human factor is larger, therefore utilizes Digital image processing technique to be classified in the floor and more can effectively guarantee the accuracy of floor classification.

Because the floor color is reflection floor surface vision and psychosensorial key character, with the evaluation close relation of classification, thereby so that in recent years certain research has all been carried out in the classification processing of color-based feature.As wear rainbow etc. by main color characteristic, utilize respectively RBF network, K nearest neighbor and arest neighbors in the neural network to the processing of classifying of wood sample image; Wang Ke very waits and utilizes even color space that the color characteristic of timber is measured and the sort research of distinguishing.

Summary of the invention

The objective of the invention is to improve classification is processed in the production run of floor automaticity and rapidity thereof, a kind of floor hierarchy classification method of color-based feature is provided, effectively raise floor production firm to the accuracy of floor category classification, both improve floor production efficiency, reduced again cost of labor.

According to technical scheme provided by the invention, the floor hierarchy classification method of described color-based feature may further comprise the steps:

(1) coloured image according to the floor sample extracts the Color Statistical feature;

(2) according to the color characteristic information of known floor sample, set up the multi-level floor of off-line sample classification according to the storehouse;

(3) when unknown classification floor sample is carried out online classification, utilize the bee-line classifying rules, ask for the similar classification of unknown floor sample according to off-line classification foundation storehouse given in advance; Utilize K nearest neighbor-bee-line method, ask for the accurate classification of unknown floor sample according to off-line classification foundation given in advance; Be specially:

(3.1) when the similar classification of unknown floor sample: calculate the distance of the color moment proper vector of unknown sample proper vector behind all similar categories combinations, and to the processing of sorting of all distances; Will with the difference of bee-line less than the Sample Maximal distance B _MaxThe similar classification of all of/4 all reduction to the classification process of next level;

(3.2) when the accurate classification of unknown floor sample: at first ask for the distance of each sample in unknown floor sample and the large class of affiliated sample, and by distance size to the processing of sorting of each sample in the large class of sample; Then add up the corresponding floor of top n distance sample class number, ask for class number maximal value k _Max, judge k _MaxWhether unique; If unique, then unknown floor is referred to k _MaxIn the corresponding classification, the classification processing finishes; If not unique, then ask for floor to be measured and all k _MaxThe distance of identical floor sample, with unknown floor sample be referred to its floor sample apart from minimum in.

Described extraction Color Statistical feature is to utilize the color moment of the coloured image of floor sample to extract the Color Statistical characteristic information of floor sample, removes luminance component to reduce illumination condition to the impact of classifying quality; And give tone component and the different weighting proportion of saturation degree component.

When setting up described floor sample classification according to the storehouse, at first determine the color moment of each floor sample according to given floor sample; Then by minimax distance classification algorithm each floor sample is carried out clustering processing, the close categories combination of color moment numerical value to together, is asked for the data mean value of similar classification floor color sample square; The minimax distance algorithm calculates ultimate range between all samples as the reference of sorting out threshold value, if certain sample to the distance of a cluster centre less than ultimate range D _Max1/2, then be included into such, otherwise set up new cluster centre.

Advantage of the present invention is: the present invention selects the color moment feature to the expression mode of floor color characteristic, compares with other expression modes, can more simple and effectively express the colouring information on floor; Utilize tone and saturation degree component to ask for the color moment characteristic on floor at the HSV color space, reduce illumination to the impact of floor classification Exact Travelling; Set up the multi-level floor of off-line sample classification according to the storehouse, improve the accuracy of floor, position sample classification; When the pairing approximation classification was judged, the approximate classification of all that will close on was included into the judgement of next stage equally, improved the accuracy of classification; Utilize K nearest neighbor-bee-line method that the unknown floor accurate classification of sample is judged, further guarantee the accuracy of floor classification.

Description of drawings

Fig. 1 is overall implementation framework figure of the present invention.

Fig. 2 is that the floor sample classification is according to the storehouse synoptic diagram.

Fig. 3 is unknown floor sample classification realization flow figure.

Embodiment

In order to improve the automaticity of unknown floor sample classification in the production run of floor, the present invention has developed a kind of multi-level floor online classification method of color-based statistical nature.Floor sorting technique algorithmic code amount is little, fast operation, high, real-time, the good stability of precision, classification effectiveness that can the unknown floor of Effective Raise sample, reduces production costs.

The invention will be further described below in conjunction with drawings and Examples.

The floor hierarchy classification method of this color-based feature, unknown floor sample is carried out the online classification implementation procedure as shown in Figure 1, at first according to the multi-level off-line of Sample Establishing floor, known floor sample classification according to the storehouse, and then to the realization of successively classifying online from coarse to fine of unknown floor sample.

Step is as follows:

1 coloured image according to the floor sample extracts the Color Statistical feature;

2 according to the multi-level floor of Sample Establishing off-line, known floor sample classification according to the storehouse;

3 when carrying out online classification to unknown classification floor sample, utilizes the bee-line classifying rules, asks for the similar classification of unknown floor sample according to off-line classification foundation storehouse given in advance; Utilize K nearest neighbor-bee-line method, ask for the accurate classification of unknown floor sample according to off-line classification foundation given in advance, recomputate at last the characteristic of approximate classification under the sample characteristic data of classification under the sample of unknown floor and the sample.

More than the 1st step specific implementation of extracting the Color Statistical feature be:

When extracting the Color Statistical feature of floor sample, select color moment to represent the Color Statistical feature of floor sample.The form of the mathematic(al) representation of three low order squares of color moment feature is:

${\mathrm{\μ}}_i=\frac{1}{n}\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{h}_{\mathrm{ij}}$

${\mathrm{\σ}}_i={\left(\frac{1}{n}\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{(h_{\mathrm{ij}}-{\mathrm{\μ}}_i)}^2\right)}^{1/2}$

$s_i={\left(\frac{1}{n}\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{(h_{\mathrm{ij}}-{\mathrm{\μ}}_i)}^3\right)}^{1/3}$

In the formula, h _IjRepresent that gray scale is the probability of the pixel appearance of j in the i Color Channel, n represents number of greyscale levels.

Select 2 chrominance properties values (tone, saturation degree) to extract its relevant colors moment characteristics and process with the classification and Detection of carrying out the floor, remove the property value of brightness.

The tolerance of similarity needs to the higher weight of tone characteristics data allocations between the floor, distributes lower weight for the saturation degree characteristic, reduces texture features to the impact of floor classification.

The weighted euclidean distance of different floors color moment vector can be expressed as

d＝ω ₁(x _H-x′ _H)+ω ₂(x _S-x′ _S)

Wherein, ω ₁, ω ₂The weights that represent respectively tone color moment feature and saturation degree color moment feature.

The 2nd step was set up the multi-level floor of off-line sample classification:

At first determine the color moment of each floor sample according to given floor sample; Then by minimax distance classification algorithm each floor sample is carried out clustering processing, the close categories combination of color moment numerical value to together, is asked for the data mean value of similar classification floor color sample square.Final formation multi-level floor sample classification as shown in Figure 2 is according to the storehouse.

The specific implementation in the 3rd step is:

When unknown classification floor sample was carried out online classification, realization flow as shown in Figure 3.

(3.1) utilize the bee-line classifying rules, other specific implementation step of Similarity Class of asking for unknown floor sample according to off-line classification foundation storehouse given in advance is:

Calculating unknown sample proper vector arrives the distance of the color moment vector behind the similar categories combination of floor sample;

To the ascending processing of sorting of all distances;

Will with the difference of bee-line less than D _MaxThe similar classification of all of/4 all reduction to the processing procedure of next level.

(3.2) utilize K nearest neighbor-bee-line method, when asking for the accurate classification of unknown floor sample according to off-line classification foundation given in advance, at first obtain k the neighbour of sample x.If maximum neighbour's number is k _Max, judge k _MaxWhether it is one; If unique, this classification ω then _MaxUnder x; If a plurality of (such as N) ask the neighbour to count k _MaxAffiliated floor sample category feature average and the distance of x:

$d_i={\left[\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{(S_i-x_i)}^2\right]}^{\frac{1}{2}};$

Wherein, S is floor sample category feature average, and n is number of features.

K nearest neighbor-bee-line method decision rule is:

Use K nearest neighbor-bee-line method to determine the affiliated accurately implementation procedure of classification in floor to be measured:

1. ask for the distance of each sample in unknown floor sample and the large class of affiliated sample;

2. by distance size to the processing of sorting of each sample in the large class of sample;

3. add up the corresponding floor of top n distance sample class number, ask for class number maximal value k _Max

4. judge k _MaxWhether unique, uniquely then carried out for the 5. step, otherwise carried out for the 6. step;

5. then unknown floor is referred to k _MaxIn the corresponding classification, the classification processing finishes.

6. ask for floor to be measured and all k _MaxThe distance of identical floor sample;

7. unknown floor sample is referred to its floor sample apart from minimum in, the classification processing finishes.

Claims (1)

1. the floor hierarchy classification method of color-based feature is characterized in that, may further comprise the steps:

(1) coloured image according to the floor sample extracts the Color Statistical feature;

(2) according to the color characteristic information of known floor sample, set up the multi-level floor of off-line sample classification according to the storehouse;

(3) when unknown classification floor sample is carried out online classification, utilize the bee-line classifying rules, ask for the similar classification of unknown floor sample according to off-line classification foundation storehouse given in advance; Utilize K nearest neighbor-bee-line method, ask for the accurate classification of unknown floor sample according to off-line classification foundation given in advance; Be specially:

(3.1) when the similar classification of unknown floor sample: calculate the distance of the color moment proper vector of unknown sample proper vector behind all similar categories combinations, and to the processing of sorting of all distances; Will with the difference of bee-line less than the Sample Maximal distance B _MaxThe similar classification of all of/4 all reduction to the classification process of next level;

(3.2) when the accurate classification of unknown floor sample: at first ask for the distance of each sample in unknown floor sample and the large class of affiliated sample, and by distance size to the processing of sorting of each sample in the large class of sample; Then add up the corresponding floor of top n distance sample class number, ask for class number maximal value k _Max, judge k _MaxWhether unique; If unique, then unknown floor is referred to k _MaxIn the corresponding classification, the classification processing finishes; If not unique, then ask for floor to be measured and all k _MaxThe distance of identical floor sample, with unknown floor sample be referred to its floor sample apart from minimum in;

Described extraction Color Statistical feature is to utilize the color moment of the coloured image of floor sample to extract the Color Statistical characteristic information of floor sample, removes luminance component to reduce illumination condition to the impact of classifying quality; And give tone component and the different weighting proportion of saturation degree component;

When setting up described floor sample classification according to the storehouse, at first determine the color moment of each floor sample according to given floor sample; Then by minimax distance classification algorithm each floor sample is carried out clustering processing, the close categories combination of color moment numerical value to together, is asked for the data mean value of similar classification floor color sample square; The minimax distance algorithm calculates ultimate range between all samples as the reference of sorting out threshold value, if certain sample to the distance of a cluster centre less than ultimate range D _Max1/2, then be included into such, otherwise set up new cluster centre.

Images