Color Image Segmentation Algorithm Based On Fuzzy C-means Clustering

As a primary way of human perception, color information is important for pattern recognition and computer vision. With the decrease of the cost for color image acquisition system and the improvement of computer processing capabilities, it is possible to deal with the high-dimensional color space. Therefore, color image segmentation has been wildly studied during last decades. Many image segmentation algorithms for gray-scale images are applied to segment every component of the color images respectively. Then the corresponding segmentations are combined to get the final result. However, without taking color information or texture information into account, this kind of algorithms cannot get a reasonable result. By mapping the color image information into high-dimensional space, cluster analysis method can deal with high-dimensional data. It is reasonable to introduce the high-dimensional information of color images into cluster analysis method to segment color images. Moreover, fuzzy theory is capable to describe the uncertainty in the data sets. Therefore, we propose an improved fuzzy C-means clustering algorithm to segment color images in this paper. Comparing with state-of-the-art, the proposed method is more adaptive and robust.(1). This paper study different type of clustering algorithms, compares their performance in real data set. The result shows that FCM algorithm is better than BIRCH, Mean Shift algorithm.(2). A new approach for color-texture image segmentation is proposed. The first problem of color image segmentation is to find a suitable color space.this paper study the performance of fuzzy C-means in different color space, evaluate the characteristics of them. We consider the Lab color space is the most suitable color space for fuzzy C means algorithm. In complex natural images, the different region often appear similar. We can effectively distinguish them by the texture. In this paper, the fuzzy C means algorithm can effectively split the target by combining the color and texture feature in complex natural images.(3). FCM algorithm requires a exactly number of cluster, and the result of FCM algorithm depend on the initialization. This paper study the methods determine the number of cluster by indicators. Though it is low efficient that use FCM algorithm repeatedly, we introduce the hierarchical methods to determine the number of cluster. And using a distance-based method to initialize the FCM algorithm. The experimental results show that the proposed algorithm is an effective segmentation algorithm.