一种用于彩色图象量化的快速聚类算法

为了对彩色图象进行有效地压缩处理，提出了一种基于模式识别技术的图象量化新算法（FSCAMMD），该算法首先把彩色图象中的颜色样本归为一类，并采用最大频度与类内最小距离最大相结合的方法选取初始类代表点--初始值优选法；然后采用欧氏距离聚类准则及重心法，求得新聚类域中心的向量值，从而得到了令人满意的量化效果。该算法不仅克服了SCA算法对聚类中心初始值选取的不足，较大幅度地减少了彩色图象量化后的总方差以及颜色失真度，而且较好地解决了重建彩色图象的整体层次与局部细节之间的矛盾，其量化效果优于SCA和其他一些聚类量化算法。

A Fast Clustering Algorithm for Color Images Quantization

A new algorithm for color image quantization based on the pattern recognition technology is proposed in this paper. First, the color samples in a color image are grouped together, and the initial representative points of the categories are chosen based upon a method of combining maximum frequency degree with maximizing minimum discrepancy, that is , an optimum seeking method of initial value of clustering center. Then both the clustering criteria of Euclidean distance in clustering analysis and the gravitational center method in mechanics are used to determine the vector values of the new clustering region centers, and the satisfying clustering effects can de obtained. This is a fast statistical clustering algorithm based on maximizing minimum discrepancy (FSCAMMD). The presented algorithm can overcome the shortcomings of the seeking method of initial value of the clustering center of SCA algorithm. Both the total mean square deviation and lack fidelity of images quantized by the present algorithm have a relatively big reduction and the effect of color image equalization is better than that of SCA algorithm and other clustering algorithms.