SIMD技术在数字图像处理中的应用研究

SIMD处理机特别适合于要求大量高速向量或矩阵计算的场合,数据缓存系统和对准网络是它的关键部件。而图像卷积是图像处理技术中最基本也是最重要的一项技术,本文根据数字图像的卷积定理对数字图像的卷积运算进行了分析,并提出了一种基于SIMD处理机的可变卷积模板的图像卷积处理器的体系结构。该处理器内部包含有接口部件、控制部件、数据缓存系统、对准电路和执行部件等。它的极高效率的数据缓存系统和对准电路成为该处理器最有特色的部分,它从根本上解决了图像卷积中的数据复用带来的CPU重复访问主存储器的问题。实现了卷积模板为3×3的图像卷积运算,从而实现了对卷积计算的硬件加速目的。最后,对这个图像卷积处理器体系结构的性能及其可扩展性进行了缜密的分析。

Research on SIMD Architecture for Digital Image Processing

SIMD processor is different from general purpose processor, which is widely used in digital and video image processing. Data buffer system and aligning network are critical parts in SIMD processor. Since digital image convolution is the most important technology of image processing. After analyzed the image convolution algorithm, the architecture of image convolution coprocessor is presented. The architecture consists of interface unit, control unit, data buffer system, aligning circuit and processing units, etc. which absolutely aims at solution of hardware acceleration while video image processing. Finally, the performance is analyzed in detail.