5. Fuzzy Clustering-Based Vector Quantization for Image Compression

The implementation of fuzzy clustering-based vector quantization (VQ) algorithms in image compression is related to three difficulties: (a) the dependence on initialization, (b) the reduction of the computational cost, and (c) the quality of the reconstructed image. In this paper, first we briefly review the existing fuzzy clustering techniques used in VQ. Second, we present a novel algorithm that utilizes two stages to deal with the aforementioned problems. In the first stage, we develop a specialized objective function that incorporates the c-means and the fuzzy c-means in a uniform fashion. This strategy provides a tradeoff between the speed and the efficiency of the algorithm. The joint effect is the creation of hybrid clusters that possess crisp and fuzzy areas. In the second stage, we use a utility measure to quantify the contributions of the resulting clusters. Clusters with small utilities are relocated (i.e., migrated) to fuzzy areas of large clusters so that they can increase their utility and obtain a better local minimum. The algorithm is implemented in gray-scale image compression, where its efficiency is tested and verified.