The Research And Implementation Of Log-Polar Transform-based Image Matching Algorithm

With the continuous development of computer technology, image matching technology has been widely used in computer vision, face recognition, photogrammetry, template matching location, resource analysis, aviation image analysis, three-dimensional reconstruction of information, missile terrain matching, scene matching, optical and radar target tracking and identification, medical image registration, image mosaic and image fusion and so on. It is an important technology in the field of modern information processing, and has important research value.Currently, in the field of image processing, people's research on the image matching technology has more in-depth, there are also many algorithms, in general, two typical algorithms are registration based on image grayscale and registration based on image characteristics. The former directly use image grayscale to match image, and achieve registration through the global optimization of some similarity measure between pixels. This method does not require segmentation and feature extraction, so it can avoid the loss of accuracy caused by image preprocessing. The latter use the method of image segmentation to extract features which reflect the shape changes of image, and as a reference feature, determining the match position through the similarity measure of the feature space. Because after the feature extraction, there is a small amount of data in this way, and features changed significantly, we can compare the differences between two images easier. The matching method based on image gray mainly take measure to scan each pixel in reference image to get difference between real-time image and reference image. This approach generally has a high rate of registration, but slowly. The matching method based on feature is not sensitive to various extrinsic changes (such as rotation, scaling and illumination intensity, etc.), it generally faster but with lower precision.As this background, through researching and analyzing on these two kinds of image matching algorithm, this paper improved the image matching algorithm based on the traditional algorithm, through the gray image binarization, Log-polar coordinate transformation, extracting the feature of coordinate projection statistics, the average absolute difference similarity comparison algorithm and other methods, and added some my own innovative methods, and integrate these parts to form a complete image matching algorithm. Finally, this paper presents the detailed design of the improved algorithm and implementation.The main research work is as follows:(1) This paper describes the research background of image matching, studies the relevant technology of image matching, and investigates the research status of image matching technology, which laid the foundation for subsequent work.(2) This paper studies the concept, sorting and the general processes of image matching, and focused on researching and analyzing on two typical methods which are registration based on image grayscale and registration based on image characteristics.(3) Through researching and analyzing on these two kinds of image matching algorithm, this paper improved the image matching algorithm based on the traditional algorithm, and gave the overall design idea of improved algorithm, including the reference image preprocessing, Log-polar coordinate transformation preprocessing, and the processing and matching of the target image, and gave the important parameters of the algorithm.(4) Finally, this paper gave the detailed design of the algorithm, including image binarization, calculating the centroid of binary image, Log-polar coordinate transformation of binary image, the statistics of the axis and angle projection in LP images, target image processing, image similarity matching, the contrast of important parameter, and gave their implementation results.