Based On The Gray Image Matching Algorithm Is Improved

The thesis consists of studying the improvement of gray-scale image matching algorithm, introducing the image pre-processing and traditional gray-scale image matching algorithms, and their investigation.Based on the background of subject and meaning of our research, the concept of image matching and the content of general image matching system are drawn.Then we analyse the principle of image matching, and classification of the technology of image matching is introduced. We mainly investigate the ongoing pre-processing technology of image matching, specially, the histogram equalization, median filtering and edge detection are discussed in detail. We compare the Robert operator, Sobel operator, Laplace operator, Prewitt operator, and Gaussian-Laplace operator with each other, experimentally. According to that, we analyse the advantages and disadvantages of each operator in their applications.Follow the discussion above, we discuss several staple and traditional gray-scale image matching algorithms, such as Multitude Sorting Relational algorithm (MSR), Fast Foourier Transform algorithm (FFT), Mean Absolute Difference algorithm (MAD), Sequential Similarity Detection algorithm (SSDA), and then analyse the advantages and disadvantages of them according to their principles. During the analysing and investigating of image matching principles, we find that there are two methodologies which can reduce the computation of template image matching algorithm. The one is to diminish the size of search space of template images in reference images, the other is to reduce the related calculation between template images and sub-images. Motivated by this, we design a new matching algorithm. At first, we need to preprocess the reference image, then sample the reference image with probability density from gray values after normalization, next we start a rough matching process, the rough matching points will be obtained by selecting proper sets of sub-images, and updating threshold dynamically. Finally, we initiate a refined matching process. By using the Least-squares method, we obtain the stationary point of quadratic curve in the neighbourhood of nine points centering matching points from rough matching process, which is the exact matching point eventually. Meanwhile, we discuss the complexity of our algorithm detailedly.In the end of our thesis, we have plentiful experiments of new algorithm, according to the comparison results of experiments of traditional algorithm, our new algorithm exhibits many advantages and powerful practicability.