The Research And Application Of Shape Matching Technology Based On Region Moment Invariants

With the rapid development of digital image technology, Shape matching technology hasgained significant attention from Academia and society, and lots of great progress has beenmade in recent years. The emergence of numerous new theories, methods and equipment hasbeen used widely in many fields, including the scientific research, industrial production,medical and health, social management, communication engineering, etc. which plays animportant role in promoting social development. This thesis focuses on shape matchingtechnology by combining region moment invariants with Euclidean distance. The key point of shape matching is to find the right way to describe the shape, and thenconstruct robust feature vector to complete the matching of similar shapes and classification ofdifferent shapes. Moment invariant is a very important shape feature indicating the overallshape of target amount, and has become a hot spot of research of shape feature as statisticallyglobal information. In this thesis at first, the theoretical and computational methods of momentinvariants are briefly introduced. The features based on moment invariants can keep theirfeatures of scale-invariance, rotation-invariance and translation-invariance. Since digitalimages dealt with are discrete, and invariant moment are defined in the2-D continuousfunction, digital error would appear and affect the final results of shape matching under thescaling changes, rotation changes and other circumstances. Therefore, based on the method ofcalculation of conventional region moment invariants, a concrete improvement method ispresented. The Hu moment invariants based on new method of calculation, to constructfeature vectors, describe the similarity between shapes well. Experiment results show that theimproved Hu moment invariants can keep stronger robustness and higher matching rate. Finally, this thesis makes systematic analysis on the LPR from image preprocessing,license plate location, character segmentation and recognition, etc. The template matchingalgorithm based on the improved Hu moment invariants method is used to the recognition oflicense plate character, then the whole process of license plate recognition is achieved byprogramming of MATLAB software, and the result is satisfactory.