基于K均值聚类的快速分形编码方法

Fast Fractal Coding Technique Based on K-mean Clustering

针对目前分形图像压缩存在的编码时间过长问题,提出了使用K均值聚类对编码过程进行加速的方法,其中聚类向量采用图像块的正规化特征向量以保证聚类的精度,并通过用部分失真搜索来完成传统K均值聚类中最耗时的最近邻搜索过程以提高聚类速度。进一步,通过结合均值图像建库、去平坦块等技巧,得到了一种快速、可调的分形编码方法。实验结果表明,相对于全局搜索,所提方法大幅地提高了编码速度和压缩比,而解码质量只略有下降。

Long coding time is the main problem in image compression based on Fractal at present, mainly due to its heavy computation of searching the best-match domain block for each range block. In this paper, a fast K-mean clustering algorithm is proposed firstly using Partial Distortion Search to replace the time-consuming Nearest Neighbor Search process in traditional K-mean clustering algorithm. Then the K-mean clustering algorithm is used to speed up the coding： scheme the domain blocks and search the best-match block for each range block in some nearest neighbors from some nearest clusters. Furthermore, by combining other techniques such as excluding planar blocks and building domain pool from an averaged image, a fast and adjustable fractal coding scheme is obtained. Experimental results indicate that comparing to exhaustive search, the proposed method improves the coding speed and compression ratio greatly with slight quality degradation of decoded image.