Research On Application Of Image Matching Based On HPC

Image matching is the foundation of many of the theory and applications in thefield of digital image processing and computer vision, and is the hot spot problems ofcurrent research. Image matching algorithm commonly has large amount of calculationand low efficiency and cannot meet some places with requirement of high real time. Therapid development of HPC provides an important way to raise the image matchingspeed, based on the Cluster platform composed of PC, using parallel computingtechnology based on MPI, the image matching algorithm parallelized can reduce thecomputing time and overall real-time performance.First of all, this paper expounds the advantages and disadvantages of a variety ofcomputer system structure used in HPC, discusses the diversification, classification ofcomputer cluster technology and typical PC cluster system, discusses parallelprocessing technology used in HPC, and introduces programming with MPI.Secondly, this paper does a deep research of Harris corner detection algorithm,Normalized Cross-Correlation method and RANSAC algorithm that are widespreadused in image matching. The RANSAC algorithm is improved based on the disparitygradient theory and Buckets technology, and its validity is proved through experiments.Finally, this paper analyses the calculation details of Harris Corner Detectionalgorithm, Normalized Cross-Correlation method and RANSAC algorithm, determinesthe parallel computing model, designs strategies of each algorithm, and realizesalgorithm through programming. A clear detailed introduction of the construction ofOSCAR cluster is showed. The experiment of the parallel image matching is done in thebuilt cluster environment.Experiments show that, the speed-up ratio of parallel image matching algorithmrunning on high performance PC cluster platform is close to the linear. Higher parallel efficiency is acquired. It proves the method in this paper is of the feasibility, the validityand real-time.