Fuzzy Wavelet-based Color Image Segmentation Using Self-organizing Neural Network

Image segmentation has been and is likely to be an important component of the content-based image acquisition and retrieval systems. This paper describes a new method for segmentation of color images. The proposed method uses two phases segmentation processes. In the 1 phase, segmentation is performed with the help of cluster validity measures and Spatial Fuzzy C-Mean (sFCM). HSV model helps in the decomposition of color image then FCM is applied separately on each component of HSV model. In the 2 phase, for fine tuning, Kohonen’s Self Organizing Map (SOM) neural network along with wavelets is used. SOM is a computationally expensive network. It has been observed that if SOM training performed on the wavelet-transformed image, then not only it reduces SOM training time but in this way makes more compact segments. The advantages of new method are: (i) it yields regions more homogeneous than those of other methods for color images; (ii) it reduces the spurious blobs; and (iii) it removes noisy spots. The technique presented in this paper is a powerful method for noisy color image segmentation and works for both single and multiple-feature data. Experiments were performed on standard color images. Experiments show better performance of the proposed method when compared with other approaches in practice.