CUDA平台的分形图像压缩方法

Algorithm of Fractal Image Compression on CUDA

考虑到分形图像压缩中,值域块与定义域块之间的匹配能够并行计算这一特点,利用计算统一设备平台CUDA进行GPU加速.提出一种GPU、CPU相结合的四叉树压缩算法,通过GPU加速最耗时的距离计算部分,而四叉树分割、初始化等部分仍采用CPU完成.在GPU加速部分,讨论了单值域块与多值域块的方法,通过分析与实验表明,后者比前者能进一步提高并行性能.与传统的纯CPU方法相比,本文的方法能够显著提高压缩速度.

In fractal image compression, the matching procedure between range blocks and domain blocks can be executed in parallel manner. Therefore, in order to accelerate fractal image compression by using GPU, we apply compute unified device architecture CUDA to it. This paper presents a hybrid quad tree compression approach of GPU and CPU, which accelerates the distance calculation that consumes time mostly in GPU side, and handles quad tree division, initialization and so on in CPU side. In GPU part, we discuss two methods, single range block and multiple range blocks. Analysis and experiments show that the latter can achieve better parallel performance than the former. When our approach is compared with traditional pure CPU ones, it can improve fractal compression speed greatly.