TW201545119A - Data clustering method applicable to color images - Google Patents

Abstract

A data clustering method applicable to color images is provided. The data clustering method applicable to color images includes: a color image is provided, and each pixel of the color image is corresponded to a color intensity value; a three-dimensional color space is constructed by color intensity value of the color image; the color space is divided into a plurality color blocks, wherein each color block is formed by selecting pixels with similar color intensity values; k color blocks are selected as an initial value of a k-means clustering, wherein k is defined as the most frequent value in the color space; splitting the color image by the initial-valued k-means clustering.

Description

Data grouping method applied to color images

A data grouping method for color images, in particular, a method for segmenting images by a k-means grouping method with a fixed starting value.

In a huge amount of information, how to find the information that can be used is extremely important. Therefore, various methods for classifying data have been proposed. Among these methods, k-means clustering is a fairly important and common method.

The basic thinking of the k-means grouping method is to select a group of k data (ie, k cluster centers) in a large number of data sets. In the k group data, the data points in each group have high similarity. Sex. Next, find the average or center of each group of data. Then, calculate the squared error function of each data point and center. How to minimize the squared error function is an important purpose of the k-means grouping method. In order to minimize the squared error function, the data points in each k group of data will be repeatedly grouped repeatedly and redistributed to other groups with higher similarity.

The main feature of the k-means grouping method mentioned above is that the calculation is simple and effective. The rate is high, so it is widely used in data mining, decision analysis, machine learning and image segmentation. However, there are still some problems to be solved in the conventional k-means grouping method. First, the initial value of the grouping is a random value, which results in inconsistent results after image segmentation, and makes the image after grouping difficult to control. Second, it can only get better results when the sample is a standard Gaussian distribution or a Gaussian-like distribution, and it cannot be widely applied to other conditions.

Therefore, in order to increase the application range and accuracy of the conventional k-means grouping method, it is necessary to improve the original k-means.

The present invention provides a data grouping method applied to color images. The three-dimensional color space formed by the color intensity values corresponding to the pixels of the color image is segmented, and the color block with the number of occurrences before the block is the k-means grouping method. Starting value. Therefore, the problem that the conventional k-means grouping method is inconsistent after grouping due to the initial value is not fixed, and the operation efficiency of the grouping can be improved.

An object of the present invention is to provide a data grouping method for color images, comprising: providing a color image, wherein each pixel corresponds to a color intensity value; constructing a three-dimensional color space, and the three-dimensional color space is colored Each color intensity value of the image is formed; selecting a plurality of pixels adjacent to each color intensity value of each pixel to form a plurality of different color blocks to block the three-dimensional color space; in the three-dimensional color space after the block The top k values with the most searches; and the fixed starting value of k-means grouping method with k value, and the color image by k-means grouping method Line image segmentation.

In the data grouping method applied to color images, the block is to combine m adjacent color intensity values into a color block of volume m ³ cubic units in a three-dimensional color space, and to make a three-dimensional color space. ^Three blocks (256/m) are formed inside. In addition, the three-dimensional color space can be an RGB three-dimensional color space.

The above-mentioned data grouping method applied to color images further includes mapping color intensity values of R, G, and B corresponding to respective pixels of the color image into an RGB three-dimensional color space, which is represented by R, G, and B. The X, Y, and Z axes form a three-dimensional coordinate system, wherein each coordinate point on the three-dimensional coordinate system represents the number of times the corresponding color appears in a particular RGB color intensity.

S101~S105‧‧‧Steps

2a‧‧‧ original image

2b‧‧‧Image of original image 2a grayscale

2c~2e‧‧‧Image segmentation of the original image 2a by the conventional k-means grouping method

R‧‧‧Red

4a‧‧‧ original image

4b‧‧‧Image of the original image 4a grayscale

4c~4h‧‧·Image of image segmentation of original image 4a by k-means grouping method with unknown initial value

4i‧‧‧ Applying the invention to color Image segmentation method for image segmentation of original image 4a

G‧‧‧Green

B‧‧‧Blue

FIG. 1 is a diagram showing the steps of a data grouping method applied to a color image according to an embodiment of the invention.

Figure 2 is a schematic diagram showing the application of the conventional k-means grouping method to image segmentation.

FIG. 3 is a schematic diagram showing the tiling of a three-dimensional color space in the data grouping method applied to color images in FIG. 1.

FIG. 4 is a comparison diagram of the data grouping method applied to color images and the conventional k-means grouping method according to the present invention after image segmentation.

The operation of the various embodiments of the present invention will be described in the following. For the sake of clarity, a number of practical details will be described in the description. However, it should be understood that these practical details are not intended to limit the invention. That is, in some embodiments of the invention, these practical details are not necessary. In addition, some of the conventional structures and elements are shown in the drawings in a simplified schematic manner in order to simplify the drawings.

Please refer to FIG. 1 . FIG. 1 is a schematic diagram of a data grouping method applied to a color image according to an embodiment of the invention. In an embodiment, the data grouping method applied to color images of the present invention comprises the following steps:

In step S101, a color image is provided.

Step S102, constructing a three-dimensional color space, wherein the three-dimensional color space is formed by each color intensity value corresponding to each pixel of the color image.

Step S103, selecting a plurality of colors adjacent to each color intensity value corresponding to each pixel to form a plurality of different color blocks to block the three-dimensional color space.

In step S104, the first k values of the most searched in the three-dimensional color space after the block are a starting value.

In step S105, the k value is a fixed starting value of a k-means grouping method, and the color image is segmented by the k-means grouping method.

In order to understand the implementation details of the above steps, please refer to Figure 2 to Figure 4. Figure 2 is a schematic diagram showing the application of the conventional k-means grouping method to image segmentation. FIG. 3 is a schematic diagram showing the tiling of a three-dimensional color space in the data grouping method applied to color images in FIG. 1. 4th The figure shows a comparison chart of the data grouping method applied to color images and the conventional k-means grouping method according to the present invention after image segmentation.

In the above step 102, each pixel in the color image is mapped to an RGB three-dimensional color space according to the color intensity values of R, G, and B. More specifically, it is converted into a three-dimensional coordinate system in which R, G, and B represent the three axes of X, Y, and Z. In the stereo coordinate system, corresponding to the RGB three-dimensional color space, the three axes of X, Y, and Z represent R (red), G (green), and B (blue), respectively, and their sizes are expressed as color intensities from 0 to 255. value. Each coordinate point on the coordinates is expressed as the number of times the corresponding color appears in a particular RGB color intensity value.

In step 102, the form of the three-dimensional color space is not limited. In the present embodiment, a three-dimensional color space composed of three primary colors of light (R), green (G), and blue (B), which are generally known, is described. Other color space systems such as HSV, CIELab, YCrCb, and CIELuv have a three-domain color space system that can also be applied to the data grouping method applied to color images disclosed in the present invention.

The k-means grouping method performs grouping of a group of data several times in a repeated iterative manner to obtain the final required data, and each iteration needs to select a starting value as a clustering center and perform data regrouping. In the conventional k-means grouping method, the initial values of each iteration are randomly generated. For example, when k-means grouping method is used to segment an image, the image is first represented by a plurality of data points ( For example, a pixel point), and then initialize k initial values in a random manner (assuming that k groups are to be divided), that is, the starting value is not fixed. And the starting value of each such grouping is a random selection method, which will lead The result of each image segmentation is greatly different. Please refer to FIG. 2, which shows a schematic diagram of applying the k-means grouping method to image segmentation. In Fig. 2, 2a is the original image to be image segmented, 2b is the grayscale image of the original image 2a, and 2c~2e is the image segmentation after the image 2a is divided into several groups by the conventional k-means grouping method. result. It can be seen from Fig. 2 that the difference between 2c and 2e is extremely large, resulting in inconsistent images after each grouping, which will increase the data grouping error and limit the application of the conventional k-means grouping method to a large extent.

An object of the present invention is to solve the problem that the starting value of the conventional k-means grouping method is not fixed. In the RGB three-dimensional color space constructed in the above step 102, the method of tiling the RGB three-dimensional color space is applied and In the block color block, the color block (k≧2) of the first k times is selected as the concept of the k-means grouping start value to unify and optimize the final image segmentation result. The k value (the number of clusters or the center of the cluster) can be freely selected by the user.

In step 103, in order to reduce the effect of data grouping operation time and removing unnecessary subtle colors, the RGB three-dimensional color space is lumped. In a three-dimensional coordinate system formed by three axes of XYZ, m adjacent color intensity values are combined into one color block (quadratic) having a volume of m ³ cubic units. Thus, the RGB three-dimensional color space to the overall three-dimensional coordinate system built configuration, i.e. the formation of (256 / m) ³ th color blocks. For example, in the coordinate axis XYZ (R, G, and B), a color block is formed by 32 adjacent color intensity values, that is, each color block of 32 ³ cubic units; and the RGB three-dimensional color space has (256/32) ³ = 512 color blocks, as shown in Figure 3.

In step S104, the first k values of the most frequently searched in the three-dimensional color space after the block are a starting value. In detail, for the RGB three-dimensional color space after the block, select the block of the previous k times (added by m ³ adjacent color intensity values) as the starting value of the k-type group, Consistently and optimize the final image segmentation results. That is, in the three-dimensional color space, the top k values are searched for the most. In one example, among the (256/m) ³ blocks, the value of the previous k large (or more) is selected as the starting value. Since we have previously diced the three-dimensional color space, the k selected starting values should be considered even if they are located in adjacent color blocks.

Referring to FIG. 4, FIG. 4 is a comparison diagram of the data grouping method applied to color images and the conventional k-means grouping method according to the present invention after image segmentation. As shown in FIG. 4, it is assumed that the image of the original image 4a is to be image-divided, and the image of the original image 4a is grayed out as 4b. At the beginning, according to the characteristics of the original image 4a, the number of clusters k (k type) conforming to the original image 4a can be freely selected, and at this time, k=3 (that is, divided into a red body, a green leaf background, and the rest of the sky background) is selected. Then, in the 512 blocks in the block-shaped three-dimensional color space (refer to FIG. 3), the first three color blocks are selected as the fixed starting value of the k-means grouping method, even if The first three color blocks are adjacent to each other and should still be selected. Then, the original image 4a is image-divided by the k-means grouping method, and each time the group is re-grouped, the first three color blocks (k=3) are used as the starting value. Performing several times of grouping in such a fixed initial value, the results obtained as image 4I are compared to those of the k-mean grouping method in which the conventional starting value is not fixed. Image 4c~4h, it can be seen that 4c~4h all show inconsistent results.

In summary, the present invention discloses a data grouping method applied to color images. The main purpose is to block the three-dimensional color space formed by the color intensity values corresponding to the pixel points of the color image, and to fix the color block with the number of times before the block is k-means grouping method. The starting value, so that the image is segmented after each image segmentation by k-means grouping. The data grouping method of the color image of the invention is applicable to various color spaces having three domains, and has the advantages of high computational efficiency and accurate calculation result.

Although the present invention has been disclosed in the above embodiments, it is not intended to limit the present invention, and the present invention can be modified and modified without departing from the spirit and scope of the present invention. The scope is subject to the definition of the scope of the patent application attached.

S101~S105‧‧‧Steps

Claims (4)

A data grouping method for color images, comprising: providing a color image, each pixel corresponding to a color intensity value; constructing a three-dimensional color space, wherein the three-dimensional color space is a color intensity of the color image Value composition; selecting a plurality of pixels adjacent to each color intensity value corresponding to the pixel to form a plurality of different color blocks to block the three-dimensional color space; searching in the three-dimensional color space after the block The top k values with the highest number of times; and the fixed starting value of the k-means grouping method, and the image segmentation of the color image by the k-means grouping method. The data grouping method applied to a color image according to claim 1, wherein the block system combines m adjacent color intensity values into a color block having a volume of m ³ cubic units in the three-dimensional color space. . While the order is in the three-dimensional color space is formed (256 / m) ³ th color blocks. The data grouping method applied to a color image according to claim 1, wherein the three-dimensional color space is an RGB three-dimensional color space. The data grouping method applied to the color image according to claim 3, further comprising: the color intensity of R, G, and B corresponding to each pixel of the color image. The values are mapped into the RGB three-dimensional color space, and the R, G, and B color intensity values correspond to the three axes of X, Y, and Z to form a three-dimensional coordinate system, wherein each coordinate point on the three-dimensional coordinate system is represented by Of a particular RGB color intensity value, the number of times that the corresponding color appears.