The Research And Implement Of Color Image Segmentation Method

In recent years, with the development of the computer science and technology, the color image has been used more and more, because the color is a powerful description element, it usually can be used to simplify the distinction of target and the process of abstracting target from scene. We can recognize more than thousands of color tones and bright degree, but only can distinguish several ten kinds of grey lays , so the color image can usually show the vision characteristic and the rules of recognition. The research of the color image segmentation has become the hotspot of segmentation field. In the paper, the color image segmentation alogrithm has been designed for the color image. In order to complete the segmentation of the color image, the system is divided into the color quantization phase, color cluster feature phase and the region-growing phase. In the color quantization phase, two schemes have been adopted, one is general palette quantization alogrithm, other is color difference-based quantization algorithm. Which scheme should be choosen depends on the size of image and the users' demands for time and qualities. In the color clustering phase, because the fuzzy cluster method is sensitive to the cluster centers, the paper provides the self-adapting fuzzy cluster method to ascertain the cluster centers and cluster counts. Not only in theory but aslo in the experiments, the paper shows the method' s rationality. Besides, the Euclidean distance is modified by use of the HSI model to improve the veracity of color cluster. Finally, considering the consecution of the district of the pixel space, on the basis of the result of color clustering, the region-growing is wanted to be processing.At first, the thesis introduces the present condition of the research of color image segmentation, analyzes the feathers of the color image, introduces several colorful spaces and summarizes their advantages and disadvantages. Next, the algorithms combine the merits of several colorful spaces and the evaluation is given, finally the experiments are done to show the segmentation system validity.