Color Image Segmentation Algorithm

Color images contain a wealth of information on color, and they are the vivid imitating and description of the object similarities. In recent years, more and more color images have widely been used in scientific research, military technology, industrial and agricultural production, medicine, astronomy and meteorology, and other areas. Color is an important element to describe image information, and according to it, we can distinguish and distill the targets that we need from the scenes. In this study, the image segmentation means the distinction among different areas with special meanings in the images, these areas are not crossed one another, and each area meets its specific consistency. With the development of the computer vision technology, color image segmentation research has become a hot spot. This study put forward a complete set of algorithms which include three stages: color quantizing, clustering and regional growth of the color images. For the color quantizing, this study puts forward two methods: one is based on the palette quantizing; another is the quantizing with combination of the smallest chromatic aberration, and one of the both needs choosing based on the size of the image and the algorithm efficiency. For color clustering, this study improves the fuzzy C-means clustering method and makes it use the European-style distance of HSI space in the situation of low illumination in order to enhance the clustering accuracy, and this study adopts the optimized fuzzy clustering to automatically select the initial clustering centers and the number of clustering. While the areas growing, by deciding the continuity of the clustering of the results of the pixel space in the areas, the study determines whether to adopt the algorithm of area growth so as to result in the best image segmentation.This thesis first introduces the background and the significance of the study, indicates the important position of the image segmentation in the correlative projects as well as its application fields, and the existing theories and methods of color image segmentation and points out their scope of application and their advantages and disadvantages. Then, in each stage of the algorithm, the study fully integrates the advantages of several color spaces, and gives the evaluation. At last, this study verifies the effectiveness of the algorithms through experiments.