Abstract: To overcome the weakness of algorithm complexity and less robust by traditional binary feature extraction algorithm, this paper presents a novel binary feature extraction method with certain global reference and robustness. Firstly, Harris feature points and Saliency map are obtained from the input matching image. Secondly, the image space is partitioned into grid in order to get the grid center, which is sorted according to the saliency degree of the corresponding grid to generate the global reference vector. Finally, the feature points are compared with the global reference vector to generate the binary feature based on the saliency weighted. Experimental results show that the proposed algorithm can greatly reduce the computational burden, and has a good ability of characterization.