An PCB Infrared Image Registration Method Of Wavelet Domain Adaptive Scale And Improved Hausdorff Distance

Image registration is a very important step in PCB infrared image detection, so the research on infrared image registration have important significance. Infrared image of circuit board have some characteristics, such as much noise and chip's heating area cross, the existing image registration algorithms are aimed at visible light image or medical image which are difficult to do infrared image registration, therefore the traditional image registration algorithms should be optimized and improved to adapt to the extraction task of circuit board infrared image registrationFirstly, introduces the present situation of the infrared image registration, several frequently-used registration methods were analyzed, finally confirm using the edge characters as matching basis and using the Hausdorff distance as the similarity measurement that completing the registration. Secondly, during extracting the edge character, deeply analyze the multi-resolution characteristics of wavelet by comparing many kinds of edge detection operator, determine to do the hierarchical multiscale edge detection by wavelet transform. Using circuit board Infrared image acquisition from laboratory to do the edge extraction simulation, analyzing and comparing the advantages and disadvantages of the extracted edge image by various algorithms. and make more perfect revision on the registration method of conventional Hausdorff distance by comprsing the various parameters of modified Hausdorff distance template matchingFinally based on these characteristics of PCB infrared image, such as the background region has single gray scale, fuzzy contour of heating components position, much noise and low contrast, combined with the above analysis and comparison,develop a image registration method that combining wavelet domain adaptive scale with improved Hausdorff distance. Firstly, extract the edge curve of the reference and the image to be registered by using the wavelet domain scale adaptive edge detection algorithm; then improve the method of traditional Hausdorff distance, segment the edge curve were and establish the basic unit for matching; coarse match the matching element of the reference edge image and the edge image to be registered, compress the search space, then do fine match to find the optimal transformation at last. Finally, complete the accurate registration of infrared image by the MATLAB simulation. The experimental results show that this method makes a marked increase in the registration accuracy, but also improved the stability and real-time performance of registration.