Research In Color Image Segmentation Techniques

Image segmentation, with certain degree of difficulty and complexity, has always been regarded as an important technology in digital image processing field. Many researchers and mathematicians have nade great efforts to develop effective methods for the achievement of satisfactory segmentation results. However, no universal segmentation method has been found so far and most of the past research is based on monochrome image segmentation. People have been striving for new potential methods which can be simpler, better and more universal.With the improvement of computer processing speed, the rapid development of the Internet and the needs of real life, color image segmentation arouses greater concern, because it is always a difficult task for people to spot the exact image thery are interested in among large numbers of images they have to deal with. Image retrieval is based on color image segmentation, therefore, color image segmentation is chosen as the research objective in this paper with the hope that some help can be provided to solve the above problem.In this paper, we present an automatic seeded region growing (SRG) algorithm for color image segmentation. The method uses regions rather than pixels as the seeds of SRG The architecture of the algorithm can be described as follows. First, the input RGB color image is transformed into HSV color space. Second, we use watershed segmentation to initialize the image. Third, the initial region seeds are automatically selected according to two rules advanced by us. Fourth, the color image is segmented into regions. Finally, region-merging method is used to merge similar or small regions. Compared with pixel-based SRG algorithm, our method can yield more robust and precise results. Experimental results have also shown that our algorithm can produce excellent results.It is feasible to combine clustering algorithm with automatic seeded region growing algorithm by applying the former algorithm before practising the latter one. In this way, better segmentation results can be achieved.