结合分类方法的并行分形图像编码算法研究

Research on Parallel Fractal-Image Coding Algorithm Combined with Classification Method

针对分形图像编码计算密集的特点,建立编码步骤的串行-并行转化机制,利用计算统一设备架构CUDA的单指令、多线程执行特性,建立分形编码在图像处理器GPU上的并行计算模型,将耗时量较大的搜索最佳匹配块的串行执行过程并行化处理,并在此基础上结合方差法对值域块进行分类以减少搜索次数.实验结果表明,该文算法与原始算法相比可达到1 200多倍的加速并保持较好的解码图像质量,满足了实时编码的要求.

Directed against the characteristic of computational intensity of fractal image encoding,a serialparallel transfer mechanism is built for encoding procedures.By utilizing the properties of single instruction and multithreading execution of compute unified device architecture（CUDA）,the parallel computational model of fractal encoding is built on the graphic processor unit（GPU）in order to parallelize the considerably time-consuming serial execution process of searching for the block of best match,on which base to classify the blocks of range in combination with the variance method in order to reduce the frequency of search.The experimental result indicates,the algorithm in this paper,as compared with the original algorithm,can achieve an acceleration of 1 200 and more times and keep the decoded image in good quality,which addresses the demand for real-time encoding.