Multi-Resolution Color Image Segmentation Based On Fuzzy Clustering

Image segmentation is the basis of image analysis, recognition and understanding. It is a key step from image process to image analysis. In recent years, as color images provide more abundant information than gray images, the scholars have attached importance to color image segmentation which becomes a hot topic in the field of image process. In this paper, new algorithms based on wavelet transform and fuzzy clustering are proposed. The main work and contribution in this paper can be listed below:1. Multi-resolution, which is one of characteristics of wavelet transform, is applied to the implementation of color image segmentation that can reduce computation and improve the computation speed. Firstly, fuzzy clustering is performed at the lowest level to obtain coarse image segmentation. Then the results are transferred to the higher level to get refined segmentation.2. A novel FCM algorithm based on Mean-Shift and cluster validity is proposed in this paper. Firstly, Mean-Shift is used to implement coarse image segmentation and initialize cluster centers of Fuzzy C-Mean algorithm (FCM). Secondly, define the optimal cluster number by the cluster validity index. Experiments demonstrate that color image segmentation based on this method and wavelet transform has faster speed and can perform better than traditional FCM algorithm.3. A new clustering algorithm based on kernel function and density function is presented, which is used to initialize cluster centers of Fuzzy Kernel C-Mean algorithm (KFCM). After combining KFCM and wavelet transform, we propose a novel fast color image segmentation algorithm. Experiments demonstrate that this method can perform faster and better than traditional FCM algorithm. Comparing with the first method, it increases computation a little bit, but the segmentation results are better and the algorithm is immune to parameters.