Color Space Subdivision-based Color Image Segmentation Algorithm

With the rapid development of computer technology, image processing has widespread application in our daily life. Image segmentation in image analysis which be taken as a crucial step serves as a focus question in the research of image techniques. Image segmentation partition an image into different non-intersect regions with some specific properties. The quality of segmentation will affect the subsequent result directly. Image segmentation plays an important role in image research. For years, image segmentation has been widely concerned and lots of feasible algorithms have been presented in the literature. Most of the old segmenting methods applied to grayscale images. Color image provides more abundant information than gray one. Visual perception of eyes for color image is more sensitive. So color image segmentation has attracted broad attention. While the old segmenting methods cannot be applied to color image segmenting directly, so researchers are devoted to the methods for color image segmentation.This thesis first analyzes the characteristics and the transformation methods of color spaces which have recently appeared in color image segmentation field, we also make a contrast between linear color-space and non-linear color-space. Secondly, we classify the mainstream segmentation into many categories:methods based edge, methods based on region, methods based special theories and so on. We studied principles and compared the advantages and disadvantages of them. Basing on these the concrete works are as follows:(1) Thought of the advantages of RGB color space:it needn't to change color space, so it avoids losing original information during non-liner transformation. It also decreases the amount of computation, the detection result will not distort original color information. So we choose RGB color space to detect. Dividing the RGB color space, and defining color information quantity for a point in RGB space according to the result of the division. To determine a pixel an edge point by computing the information quantity of the pixel in its neighborhood. In this way, the method determining a point an edge point takes into account the correlation in each color component to a certain extent, which translates the computing from victors to scalars in a natural way. The method in this paper is novel that differential to the traditional way, and it can detect more distinct edges.(2) L*a*b* color space is a uniform color space and can calculates the difference of human vision more accurate. Any color in natural world can be expressed in L*a*b* color space. A new approach of edge detection based on the measure of color information is proposed with the analysis of L*a*b* color space. Structure cubes in L*a*b* space. Define color information measure for a pixel according to the change of volume to determine whether a pixel is an edge point. This method used the information of luminance and chrominance of color image synthetically. The experiment result proves that this method can detect color edges fast and efficiency and be superior to the traditional ones.(3) According to the two methods of edge detection, we implement color image segmenting by edge connection technology. We label each edge point and connect these similar pixel points according to criterion, then form a continuous edge. The experiment results proved that this segmenting method has better segmentation results than old ones.