| Image matching is a classic problem in image processing, and it is extensively applied to the field of computer vision, pattern recognition, medical image processing, etc.In this paper, in order to improve matching speed and robustness, we deeply study on the two aspects about matching algorithm, which is matching similarity function and fast template searching strategy, based on the analysis of related works. The research includes:1) In the research on matching algorithm based on image gray value, firstly, we make an improvement on fast normalized correlation, and propose a method of differential ordinal array based on differential and recurrence sum formula. Secondly, K-value template method which reflects the main character of matching object instead of original 256-value template is proposed. Then, the two methods mentioned above are combined with to reduce the computation in matching processing. Thirdly, fast coarse-fine searching method based on k-value template is adopted, and the cost time is improved two orders of magnitude in contrast to the current correlation algorithms.2) In the research on matching algorithm based on image character, a matching algorithm for a certain angle rotation invariant is studied, based on interesting area weighting. Meanwhile, an improved algorithm for a rotation invariant template matching method base on circle projection is also researched. By fast coarse matching based on mean-residual circle projection, computational complexity is reduced and accurate template matching is performed only on the matching candidates.3) A design idea for fast algorithm of image rectangle window weighing processing and programming skill is concluded, in order to further improve processing speed of the whole matching system.Experimental results show that the developed algorithm is robust and fast under the condition of un-linear changing of local lighting, noise, target matching of irregular shape and even matching in complex background and rotation invariance. |
PDF Full Text Request