Design And Performance Optimization Technology For Digital Image Matching System On Very-Long Instruction Word Processor

This paper mainly introduces optimization technology of image processing algorithm based on the very-long instruction word processor. These optimization techniques fall into two categories: storage optimization and ILP optimization. Storage optimization is taking use of DSP multi-level storage architecture, characteristics of algorithms and the data I/O scheduling to reduce runing time of program. Instruction-level parallelism is based on architecture of DSP , launching multiple instructions in a cycle to reducing running time.In order to verify algorithm, a digital image matching system which is based on TI TMS320DM642 DSP is realized. The software is divided into two parts: the software in PC and the software in DSP. The software in PC realizes two functions: First, opening the picture and transfer useful data to the DSP, and the second is to receive the results of DSP and display results. The function of Software in DSP is to process data from PC by instructions, including image filtering, edge detection, binarization and picture matching.There are some contributions in this paper.Applied many storage optimization methods to sliding window algorithm of image processing. Experiments proved that the optimum ratio of the acceleration of 30-70..Optimized distance transform algorithm with ILP. In accordance with the software optimization flow in C6000, using many ILP optimization methods such as writing linear assembly, reduce redundancy ,using the instructions which can accessing longer data , using SIMD instructions, software pipelining ,modulo scheduling and so on.Several storage optimization method proposed to optimize a few steps of QR decomposition algorithm respectively. After analysis, to the 512×512 matrix, the data I/O of matrix tanspose can reduce by about 15%. The data I/O of Triangulation can be reduced to 1.9%.Realized the image matching system with optimization method above,.