KR0126447B1 - Color image split-sampling method - Google Patents

Abstract

A dividing and extracting method of natural color image comprising : the step condensing detail sectors having same color property to use the shortest neighboring method among multi-variable collective analysis method; the step memorizing the values of said detail sectors in temporary memory and condensing them; the step making the N*3 data set used as the input of the multi-variable collective analysis method from said temporary memory; and the step extracting sub-sectors from the temporary memory after said data is assigned to the temporary memory; is disclosed. Thereby, sub-sectors having similar color property are automatically divided and extracted from natural color image by using computer system.

Description

Color Image Segmentation Extraction Method

The present invention uses a computer system to automatically sub-regions having similar characteristics of color from color images displayed on a screen of a computer that plays a major role in the multimedia era. It is related to a method of dividing by extraction.

In general, when the product development in the advertising industry, computer game industry, image industry, cartoon industry, printing industry, textile industry, etc., color images are edited. In this case, the process of dividing the color image into sub-regions having similar characteristics of color is indispensable.

The reason for this is that each sub-area obtained by dividing and extracting a color image is displayed on a computer screen by reconstructing or rearranging the pixels, synthesizing these sub-areas with another color image, or constituting a sub-region from each sub-region. This is because extracting color feature information of pixels, or extracting form feature information from the contour of each sub-region is a very important part of the color image editing process.

Up to now, the industrial sites exemplified above have segmented the color images using a computer system with image editing function. At this time, the pixels of the image obtained by enlarging the image are observed one by one, and the image is segmented by hand.

Therefore, the first is not precise, the second is taking a lot of time. This not only increases the cost of product development but also causes the quality of the developed product to deteriorate.

These problems are directly related to the cause of the deterioration of the international competitiveness of Korean goods, so it is urgent to develop a method that can be automatically solved. Developing this method requires simultaneous understanding of three theories: understanding the theory of color science, understanding theories of image processing, and understanding advanced theories of statistics. For this reason, the development of this method has not been attempted in Korea until now.

In the process of automatically segmenting and extracting sub-regions having similar characteristics of color from the color image using a computer system, the following matters are considered as important.

First, the subregions extracted by dividing should be accurate to match the human visual judgment.

Second, it minimizes execution time required in the process of dividing and extracting. The case where this processing time is important in terms of area is the case of an interactive system by a computer.

The methods of color image segmentation and extraction that have been studied and published in the academic field of the United States so far are one-dimensional or two-dimensional histograms based on color features (for example, CIE L * a * b * described below). ) Is a segmentation method that finds a peak at a peak and repeats it.

However, the problem of color image segmentation and extraction is not a problem to be dealt with in one or two dimensions. Because the components of the color image displayed on the computer screen are red, green, and blue, and the division of the color image is in a three-dimensional space consisting of these three elements. Because it must be done. When color image segmentation is not performed in three-dimensional space but in one or two dimensions, color image segmentation may be inaccurate.

Until now, if color image segmentation is not performed in three-dimensional space of CIE L * a * b transforming red, green, and blue, but in one or two dimensions, color segmentation is not accurate. You get

Until now, the fundamental reason why color image segmentation has not been achieved in the three-dimensional space of CIE L * a * b, which transforms red, green, and blue, is the memory required in the process of segmenting color image in three-dimensional space. Since the size of is larger than the capacity of the main memory of the central processing unit (CPU), the reason for this is expedient (the specific reasons for this will be described in detail below).

In digitizing the above-mentioned color characteristics, the RGB space (red, green, and blue three-dimensional color system) consisting of red, green, and blue, which is a component of a natural color image, is applied to data digitized with RGB. Mutual comparison between colors is impossible. Because of these problems, the International International Organization for Color (CIE), the International Commission on Illumination (CIE), in 1976 transformed the RGB space and released a three-dimensional space called CIE L * a * b *. Since the CIE L * a * b * space is accurately compared with each other by digitized data, this CIE L * a * b * space is used worldwide for segmentation of color images. In the standard KS, this CIE L * a * b * space is adopted as the industrial standard.

Color image segmentation is performed manually by using a computer system with image editing function and by using one-dimensional or two-dimensional histogram, which cannot be done in three-dimensional space by the characteristics of color, that is, CIE L * a * b *. Problems of the background achieved by the legal method are important motivations of the present invention.

The problem that occurs in the process of dividing a natural color image in three-dimensional space by the characteristics of color is the multivariate cluster analysis method, which is a method of the multivariate analysis method of statistics. This results in the first problem encountered during the application of. The following are the area problems caused by applying this method to color image segmentation.

When the multivariate colony analysis method is applied to color image segmentation, a distance matrix is created during the first calculation process. In this case, if the number of entities input to the calculation is N, a matrix of size N × N is created.

For example, in the case of a small sized color image composed of 300 x 300 pixels as shown in FIG. 1, the number of initial entities is 90,000 (= 300 x 300). In this case, the size of the distance matrix is 90,000 x 90,000, which requires the capacity of the main memory device to hold 8,100,000,000 numbers.

Given that the capacity of the main memory of the supercomputer (CRAY C90), which is the highest among the existing computers, is 250,000.000 bytes, the multivariate colony analysis method is a small-sized color image of the above two cases. In the first calculation process of dividing them, the calculation is interrupted due to the problem that the required memory size exceeds the capacity of the main memory. This is a fatal problem.

This fatal problem is the obstacle, so that all the color image segmentation so far cannot be applied to multivariate colony analysis in three-dimensional space, and the segmentation method is repeatedly performed by finding vertices in one- or two-dimensional histogram. Has been applied.

The problem with this method is that precision is inaccurate and the processing time is enormous. Therefore, when researching a new method in a university laboratory, it is used only as an experiment and there is no case of use in an industrial field.

In view of such a realistic situation, an object of the present invention is to provide an interactive method capable of accurately and quickly segmenting and extracting a color image by applying a multivariate colony analysis method regardless of the size of the color image.

1 is a conceptual diagram of the size of a distance matrix generated from a 300 × 300 color image.

2 is a flowchart of a method of segmenting and extracting a color image using a temporary storage method.

Referring to the configuration and operation of the present invention for achieving the above object in detail as follows.

First, the structure of temporary storage, which plays a key role in the segmentation and extraction of color images, and the function of each item will be described with reference to the table.

When visually observing the color image displayed on the computer screen, it can be seen that it is mostly composed of one or more sub-areas. When one observes each sub-area in detail or enlarges it by using a computer system that edits a color image, it is composed of one or more minute regions formed of a group of pixels having the same color characteristics. It is easy to see. In view of this fact, the present invention invented the temporary storage shown in [Table 1] created in the immediately preceding process of calculating the color image segmentation by applying the multivariate colony analysis method.

[Table 1] Temporary Storage

The temporary storage of the above [Table 1] consists of six items.

The first item is Clusterr I. D. This is a number assigned to each cluster in the sub area.

The second item is I. D. This is because the number assigned to the ID item, which is the second item of the temporary storage area corresponding to the minor region formed by the group of pixels having the same color characteristics, is the color image segmentation process. It is not changed until after it is finished and recorded in the final result and file.

The third item is the Pixel Value. This is a pixel value used when displaying on a computer screen pixel by pixel.

The fourth item is L *, a *, b *. This is a numerical value corresponding to the three characteristics of the color obtained by converting the R (red), G (green), and B (blue) values of the pixel, that is, coordinates indicating the position of any point in the CIE L * a * b * space. admit.

The fifth item is N.O.P. This is used to display a color image of a detailed region on a computer screen as the (x, y) coordinates of each pixel having the same pixel value, or to derive information about a position on the computer screen from each pixel.

Next, we propose an algorithm for the color image segmentation method using the temporary neighbor method and applying the nearest neighbor method among the multivariate cluster analysis methods. A description with reference to FIG.

In step 100, a file of color images consisting of m × n matrices is read. Where m and n represent the number of rows and columns of the matrix, respectively. Each element of this matrix consists of pixel values, R (red), G (green), and B (blue) values required for display on a computer screen.

In step 101, a temporary storage is created using the matrix. In cluster I.D., the first item of temporary storage, 0 is set as the initial value. From the upper left of the matrix, the pixel values of each row are sequentially checked, and each time a new pixel value is found, the first Arabic numeral 1 is inserted into the ID of the second item of the temporary storage. Whenever the details change, insert a new Arabic numeral. The newly found pixel value is inserted into the pixel value, the third item of the temporary storage. In L *, a *, and b *, the fourth item of the temporary storage, the L *, a *, and b * values obtained by converting the R (red), G (green), and B (blue) values corresponding to the pixel Put it in. In the fifth item of the temporary storage, the same pixel value is put in N. O. P., and the number of pixels having the same pixel value is inserted. In the sixth item of the temporary storage, the (x, y) coordinates of pixels having the same pixel value are inserted. The number of temporary stores created at this stage is N. Here, N is equal to the number of subregions having different color characteristics constituting the color image.

In step 102, N sets of L *, a *, and b * data are extracted from each temporary store according to the order of the ID of the second item of the temporary store, and the N × 3 matrix is obtained as shown in Table 2. Make.

Table 2 N × 3 Matrix

In step 103, L ×, a *, b * data of N × 3 are input to create a distance matrix in the form of N × N symmetry from a set of N entities. Here, the distance between each entity is calculated by the Euclidian distance calculation method.

In step 104, the minimum distance is found in the distance matrix created in the previous step.

In step 105, cluster the clusters having the shortest distance to form a new cluster, label the colon and then move the entities in the distance matrix to the following steps 106, 107, Update in step 108.

In step 106, the rows and columns associated with the colonies found in step 104 are deleted from the distance matrix.

In step 107, the shortest distances between the newly formed colony and the remaining other colonies are calculated. The number of shortest distances calculated here is equal to the number of colonies remaining.

In step 108, a column and a row of newly calculated shortest distances are added to the distance matrix.

In step 109, I.D. according to the number of colonies by I.D. is stored in a file. At this time, the contents of the generated file will be in the form shown in [Table 3]. Table 3 is an example where N is 9. In other words, when 9 pixels have different color characteristics, the ID of the left end of [Table 3] is a unique number initially assigned, and the number of colonies at the top is set from 9 to 1 from each stage. The number of each cell except the left and top is reduced to 9 to 1, and in each case is the Cluster ID indicating the same colony.

[Table 3] I.D. according to the number of colonies by I.D.

In step 200, it is checked whether all colonies are bundled into one colony, and if the number of colonies is larger than 1, the process returns to step 104 and performs each step sequentially from step 104 to step 200.

In step 201, the user visually observes the color image displayed on the computer screen and inputs the number of different kinds of colors constituting the color image using a computer key board.

In step 202, the same number as the number entered by the user is found in the number of colonies at the top of the file generated in step 109, and the numbers in the column corresponding to the number are located at the left end of the file. Extract in order of ID.

In step 203, each element (here, referred to as a number) of the column vector extracted from the colony analysis result in step 202, which is the step immediately above, is placed in the order of the ID of the second item of the temporary storage created in step 101. Therefore, we put it in the Cluster ID, the first item of the temporary repository.

In step 204, the first item of the temporary storage, the value of the Cluster ID is searched for and displayed for the temporary storages having the same value. In step 205, the result of examining whether the divided color image matches the visual judgment of the human being If does not match the visual judgment, steps 201 are performed sequentially from step 201 to step 205. Finally, the result obtained is a collection of temporary repositories whose values of Cluster I.D., the first item of each temporary repository, have the same value, so that visually similar sub-regions are obtained.

As described above, according to the method of extracting color image segmentation of the present invention, first, inaccuracy and time wastage elements in the process of segmenting and extracting color image image using a computer system having a function of editing color image are solved. Can be.

Second, by inventing a temporary storage method that can directly apply multivariate colony analysis method, rather than a convenient method to find and repeat vertices in a one-dimensional or two-dimensional histogram based on color characteristics. Until now, due to the excess capacity of the main memory of the computer, it was possible to segment the color image by the multivariate colony analysis method which was impossible anywhere in the world.

Third, the time required for finding and performing vertices in a one-dimensional or two-dimensional histogram, which has been used as a shortcut, has been shortened by 300 to 500 times.

Fourth, the method of segmenting and extracting the color image by finding the vertices in a one-dimensional or two-dimensional histogram, which has been used as a shortcut, has been irrelevant to the user's intention. There is a problem with this. However, the method of the present invention can obtain a desired result through an interactive method between the user and the computer system in the process of segmentation and extraction.

Fifth, in view of the overall performance and functional aspects of the present invention described above, the process of editing color images during product development in the advertising industry, computer game industry, video industry, cartoon industry, printing industry, textile industry, etc. In the case of using the method of the present invention, it is easy to develop excellent products, can reduce the development cost or there is an advantage that can make a significant contribution to strengthen the international competitiveness of our products.

Claims (2)

The method for extracting color image segmentation, characterized in that the temporary storage that plays a key role in the process of segmenting and extracting the color image comprises six items as follows. (1) Cluster I.D. assigned to each colony corresponding to the sub area, (2) I.D. of the temporary storage corresponding to each sub area formed of a group of pixels having the same color characteristics. ID assigned to each item, (3) pixel value used to display in pixel unit on computer screen, (4) color of R (red), G (green), and B (blue) (3) L *, a *, b *, (5) The number of pixels with the same pixel value (NOP the number), which is a numerical value corresponding to the coordinate representing the position of a point in the CIE L * a * b * space of pixels), (6) When displaying the color image of the detail region on the computer screen as (x, y) coordinates of each pixel having the same pixel value, or deriving information about the position on the computer screen from each pixel. (X, y) coordinates to be used. A method of segmenting and extracting a color image using a temporary storage according to claim 1 and applying a multivariate colony analysis method, the method comprising: reading a file of a color image composed of m × n matrices, and N temporary storages using the matrix; Creating step 101; extracting the N sets of L *, a *, and b * data from each temporary store according to the sequence number of the ID, which is the second item of the temporary storage, and creating an N × 3 matrix; Step 103 of generating N * 3 L *, a *, b * data from a set of N entities to form a symmetrical form of N × N, and two of the shortest distances in the distance matrix created in step 103. Finding a colony 104, tying the colonies with the shortest distance to create a new colony, labeling the colony 105, and from the distance matrix Deleting the columns and rows associated with the colonies found at 104, calculating 107 the shortest distances between the newly formed colony and the other colonies, and the newly calculated shortest distances. Step 108 of adding columns and rows to the distance matrix, step 109 of storing cluster IDs according to the number of colonies by ID, and checking whether all colonies of the step are bundled into one colony, and the number of colonies is 1 If it is larger, the process returns to step 104 where step 200 is performed sequentially from step 104, and the user visually observes the color image displayed on the computer screen to determine the number of different types of colors that constitute the color image. The user enters the data using the number of colonies at the top of the file generated in step 109. Finding a number equal to one number and extracting the numbers in the row corresponding to the number in the order of the IDs in the left end of the file; and each element of the row vector extracted from the colony analysis result in step 202. (In this case, a number) is inserted into the Cluster ID, which is the first item of the temporary store, in the order of the ID, which is the second item of the temporary storage created in step 101, and the value of the Cluster ID, which is the first item of the temporary storage, is the same. Step 204 of finding and displaying temporary storages having a value; and if the divided color image matches the visual judgment of the human being, and if the result does not match the visual judgment, each step is sequentially performed from step 201. The color image segment extraction method, characterized in that step 205 is performed.