Design And Implementation Of A Novel Template Matching Algorithm Invariant To Rotation

The automatic target recognition is that the target image is extracted andmatched automatically in all view fields, i.e. this system can implement targetcapture, recognition and tracking through a serial of image data processing. Oneobjective that scholars develop machine vision is to enhance the efficiency ofindustrial production, and also machine vision could help human deal with manytough tasks that are dangerous to we human. The main arithmetic includes imagepretreatment, partition, target detecting, character computing, correlationmatching, recognition and classified, motion analysis, and so on.The template matching process involves cross-correlation template with thescene image and computing similarity between the to determine the displacement.A large number of correlation-type algorithms have been proposed. One of theapproaches is to use an image pyramid for both the template and the scene image,and to perform the registration by a top-down search. Other fast matchingtechniques use two pass algoriths; use a sub-template at a coarsely spaced grid inthe first pass, and search for a better match in the neighborhood of the previouslyfound positions in the second pass. However when objects in the image arerotated with respect to each other, the methods described above can not be used,and a set of templates at different orientations is to be used. This procedure ishardly practical for real-time processing when the rotation angle is arbitrary orunconstrained.In this dissertation, a novel fast template matching algorithm invariant torotation has been proposed to solve the problem discussed above. The methodconsists of three stages. The first step and the second step aim to extract a coarsematching point in a pyramid image, Hu moments and Ring projection are used inthese two steps respectively. After acquiring the coarse matching points, Zernike moments are applied to compute the fine matching point at the highest resolutionimage. To accerelate to matching speed further, a lookup table is set in the secondstep to reduce the computation complexity to a great extend. Finally, the phaseinformation of Zernike moments are used to evaluate the approximate rotationangle.Section2introduces the traditional template matching methods invariant torotation. Both the advantages and disadvantages of those methods are presentedin this section. Section3presents the novel fast rotation invariance templatematching algorithm. In section4, various simulation results from the use ofproposed method are reported and compared with results getting fromconventional template matching methods in terms of matching speed, matchingaccuracy, and the presence of Guassian noise and image blur. The testing resultsshow that the proposed method has a faster matching speed than Log polar FFTalgorithm and the ring projection coarse-fine matching method. Finallyconclusion is given in section5.