Adaptive Color Image Clustering Segmentation System Based On SOFM Network And Fuzzy Algorithm

The traditional clustering segmentation algorithms can produce clustering overlap easily,and result in error classification of image pixels for colorful images with complicated background, they need the color number to be predefined as a prior parameter and always run with low efficiency. The cluster number and center affect the quality of the final result directly, therefore, adaptive to find a suitable number and center has great significance and engineering application value in color image clustering segmentation.In this paper, proposed a color image clustering segmentation system based on a two-level structure, which allow the number of color and cluster centers to be determined automatically. The first level of our system employs the self-organizing feature map (SOFM) to map colors of image on a two dimensional feature map. Because the traditional SOFM often produce result with color distortion, it setting the map size with different length according to the complexity of images. The second level is to process the data produced from SOFM. In this level, fuzzy C-mean (FCM) is used to cluster color images into 2 to kmax clusters (an upper limit on the number of clusters), and Xie-Beni index is applied to select the optimal number by the minimal value. After this, the Gath-Geva (GG) method is used to adjust the cluster centers. Finally,every pixel is calculated and classified according to the distance competition and replaced by the centers.Various images with different complexity are used to test the proposed system and the experimental results show that our system can obtain the better result than traditional SOFM, K-Mean, GK and GNG; it owns both advantages of the neural networks and fuzzy clustering algorithm.