Image matching Under Varying Illumination Using Uniform Symmetric-Local Binary Pattern

In this paper, we describe a novel and robust feature descriptor that is invariant to any complex brightness changes. Firstly, the local maxima is detected with recursive DOG scale space as candidate interesting points. Secondly, the main orientation is determined statistically according to the oriented gradients histograms in a circular neighborhood around interest point. Thirdly, the 128-bit descriptor is structured statistically according to the gradient magnitude in a circular neighborhood around the feature point. And then, the 64 bit descriptor is structured by extracting the texture information of the interest point neighborhood with the proposed Uniform Symmetric-Local Binary Pattern (US-LBP). Finally, combined two kinds of texture statistics form 192 bit descriptor. 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. And this method matching accuracy is much higher than SIFT, SURF, ORB, BRISK, FREAK algorithms, especially the experiments finally show the superiority of our descriptor compared to existing ones under complex illumination changes.