Color Image Segmentation Algorithm Based On Clustering

Image segmentation is an important part of image processing technique, for the featureextraction, target identification; image understanding work provides useful information. Sincebefore the technical level and hardware level, image processing has been staying in forgray-scale image segmentation, the image segmentation method has achieved good results.With the continuous improvement of science and technology, the processing of color imagehardware equipment popularization, color image segmentation techniques becomeincreasingly prominent.The existing color image segmentation, mostly is false color image segmentation.theyignore the color information of the image, direct gray-scale image segmentation method isextended to color space. Although the gray image segmentation techniques have beendeveloped comparatively perfect, but not all the methods are suitable for extended to colorspace. Because of the theory of the algorithm, clustering image segmentation methods is verysuitable for extended to color space. This paper studies the previous methods of usingclustering algorithm,found that the method has disadvantages as follow:1easy to fall intolocal minimum.2initial clustering center influenced the outcome of segmentation.3clusternumber need to manually identified4algorithm computation is very large. According to theclustering algorithm is sensitive to the clustering center,combination the theory of particleswarm optimization.at the same time, According to the particle swarm algorithm is easy toconverge to local minimum defect, made beneficial improvement, strive to make the particleswarm algorithm to escape from local extreme bondage, enhance the accuracy ofconvergence.The paper mainly includes the following aspects:First, analyze and summarizes the advantages and disadvantages of several color spaces,choose the most suitable color image space. each component of color space as a vector, usingfor the segmentation of the set of pixels.Secondly, according to the clustering algorithm for image segmentation is too sensitiveto the initial clustering center number and position, proposed introduction of particle swarmtheory was improved, the use of particle swarm algorithm to help identify clusteringalgorithm to the initial clustering center In addition,the particle swarm algorithm is easy to fall into local extreme value,for thisproblem, propose a useful improvement,to do condition disturbance on local and globalextreme values.and proved it correct on theoretically.Finally, designed two types of experiments, evaluation experiment improved particleswarm algorithm using the evaluation function, and improve the effectiveness of thealgorithm is proved by experiments; second experiment using the improved imagesegmentation method to improve clustering segmentation method analysis, experimentalresults show that the improved algorithm is effective.