A Local Invariant Feature Description and Matching Algorithm Based on LBP

In order to solve the problem of high computational complexity and the low matching speed of the traditional SIFT algorithm,a novel image matching algorithm based on LBP descriptor is proposed.Firstly,the local maxima is detected with DOG scale space as candidate interesting points.Secondly,aiming at avoiding rotating the image,the main orientations are determined statistically according to the oriented gradients histograms in circular neighborhood around the interesting point.And then,the 132-bit descriptor is structured by extracting the texture information of the interesting point neighborhood with the proposed speedy LBP descriptor named S-LBP.The computational complexity of the descriptor is reduced remarkably with the implementation of the proposed speedy LBP(S-LBP) descriptor.Finally,the descriptor matching is carried out with the logical AND.The experimental results show that the proposed algorithm has the excellent features of scale invariance,rotation invariance,affine invariance and illumination invariance in the process of image matching.This method outperforms the SIFT and CS-LBP algorithm in efficiency elevation with the same matching correctness rate.In addition,the proposed method exceeds the SIFT algorithm in illumination invariance evidently.