Design And Implementation Of Image Matching Based Wavelet Transform

Image matching is an important technology, it plays an important role in manyaspects such as pattern recognition and so on. With the rapid development of scienceand technology, the precision and real-time performance of image matching algorithm isvery important in many fields. Therefore, how to reduce the time cost of imagematching algorithm with keeping precision is a very significant study.This dissertation focus on the study of image matching based gray value, thisresearch mainly includes the following three parts:Firstly, this dissertation researches the similarity functions and search strategieswhich used frequently. It analyzes those functions and strategies, an image matchingalgorithm(NCC) based on discrete wavelet transform is provided. It uses normalizedcross correlation and adaptive SSDA as similarity functions, the pyramid search strategyis used as search strategy. The proposed algorithm is designed with Visual C++programming language and experimented under different condition.Secondly, this dissertation proposes a architecture of image matching basednormalized cross correlation and designs it with Verilog hardware description languagein the Modelsim environment. Logic synthesis is completed by ISE10.1, debugging isimplemented with Xilinx Virtex-2XC2VP30FPGA board.Finally, a architecture of discrete wavelet transform based GPU and FPGAarchitecture of NCC for pyramid search strategy are proposed, they are designed withCUDA C programming language and Verilog hardware description languagerespectively. Function validation and logic synthesis are implemented with Modelsimand ISE10.1respectively, debugging is implemented with Xilinx Virtex-2XC2VP30FPGA board.According to the test data, the performance of image matching algorithm baseddiscrete wavelet transform is better than other reference algorithms. For meeting thedemand of real-time better, this dissertation proposes two FPGA architectures for imagematching. According to simulation results and synthesis reports of Modelsim andISE10.1, performance of two FPGA image matching architecture is improved and betterthan the corresponding algorithms implemented with Visual C++programminglanguage. When the sizes of template image and reference image are32*32and256*256respective, the FPGA architecture for image matching based NCC can finishimage matching task with13.1ms time cost, the time cost of FPGA architecture forimage matching based discrete wavelet transform is6.8ms.