An Improved Illumination Invariant SURF Image Feature Descriptor

Abstract

A speeded up robust feature descriptors and matching algorithm based on brightness order is designed to overcome the problems of precision and robustness of the original SURF algorithm. Pixels are sorted and segmented according to gray values in the features support region. By establishing an index table, each segment of the pixel is represented to form a descriptor, and each segment of the descriptor is serially connected to form a feature descriptor which is used to match the image. The experimental results show that the proposed method is more higher matching accuracy, and better robustness to linear or nonlinear illumination changes compared with SURF.