Research On Robust Deformable Template Matching Algorithm Based On Best-Buddies Similarity

Template matching is widely used in computer vision,such as target detection,target tracking,video surveillance,image stitching and so on.At present,a large number of template matching algorithms have been proposed,in which Best-Buddies Similarity(BBS)is a similarity measure for template matching.This method has good robustness and high matching precision,and can overcome a certain degree of geometric deformation,background occlusion and illumination change.However,this method still has two performance drawbacks.First,the most likely area found by this method must be the same size as the given template,so when the size of the target changes in the test image,the matching performance is poor.Second,the method scans the entire image through a sliding window to find the region with the highest BBS score,which consumes a lot of computing time.In this paper,the BBS algorithm is improved so that it can guarantee the matching accuracy and it has an ideal computing speed.The main contents and innovation points of this paper include the following aspects:1、A deformable template matching algorithm based on BBS is proposed.In this paper,a deformable template matching algorithm is proposed to solve the problem of poor matching precision when the size of the target changes greatly in the test image.The algorithm consists of three steps: proposal generation,proposal selection and BBS similarity computation.First,we use the Multiscale Combinatorial Grouping(MCG)algorithm to generate a large number of proposals with various sizes.Then we eliminate the proposals which have obvious discrepancy according to a custom template size-based screening mechanism.Finally,we calculate the BBS value between the template and the selected proposals,in which the proposal with the highest BBS score is our target.Experimental results show that the proposed algorithm is superior to the existing template matching algorithms in accuracy and reduces the computational complexity at the same time.2、An enhanced deformable template matching algorithm is proposed.In order to further improve the matching accuracy and reduce the search time,an enhanced deformable template matching algorithm is proposed in this paper.Choosing a method for proposals generation has a close relationship with the accuracy of template matching,so we use the Edge Boxes algorithm to generate proposals.In the selection process,in order to reduce the amount of calculation,we propose a new multi-layer screening scheme,that is,a screening framework based on the combination of size of the template,Normalized Cross Correlation(NCC),and color histogram.And we still use the BBS algorithm in similarity calculation.The experimental results show that the algorithm can accurately match the target in a shorter time.