Research On Algorithms For Image Segmentation Based On Improved Fuzzy Clustering

Image segmentation is the foundation of analysis, comprehension and recognition, so it’san important process and difficulty in imaging processing. Because of the ambiguity of image,some researchers introduce fuzzy theory into imaging processing and use fuzzy clustering inimage segmentation. In this paper image segmentation based on FCM is discussing and newalgorithms are proposed to solve some problems in this area.Firstly, knowledge about fuzzy theory and FCM are introduced. Then the advantages anddisadvantages about FCM used in image segmentation are analyzed. At last, new algorithmsto solve these disadvantages are proposed and some experiments are conducted to show theeffectiveness of these algorithms.This paper proposes a improved algorithm that incorporates the spatial information intothe FCM, which can overcome some shortcomings the traditional FCM has. In this algorithm,FCM will be used for initial segmentation. After that, selective smoothing will be done, whichchanges the gray value of the pixel according to the classification its neighboring elementsbelong to. Finally, the image treated will be segregated again using FCM. As the experimentalresults shows, this algorithm can suppress noise efficiently.Using FCM in image segmentation needs people to determine the clustering number. Tosolve this problem, an improved algorithm is proposed. Firstly, this algorithm executes subimage decomposition to the image based on the quad tree structure (original image is dividedinto2×2sub images equally and then the sub image is divided into2×2sub images equallyagain) until the sub image meets some conditions. Then the sub image is segmented by usingFCM with the clustering number2. Lastly, region merging is executed according to the areaof the region and the Bhattacharyya distance of the two adjacent regions’ histogram, so itcould avoid determining the clustering number directly. It could be seen from theexperimental results that this improved algorithm could achieve a good segmentation result.Due to reducing the samples’ number when executing FCM to the sub image, computationamount could be reduced to some extent.