Abstract: The traditional floating feature descriptors are in high memory load and slow in matching. To best address these problems, this paper proposed a novel binary feature descriptor based on gradient statistic information comparison. Firstly, the image patch around the keypoint is divided into sub-regions, and our binary descriptor is constructed by comparing the gradient statistic information of these sub-regions. Then, a multi-gridding and multi-support region strategy is applied to boost the discrimination of our descriptor. Finally, a simplified Ada Boost algorithm is applied to realize the descriptor dimension reduction. The experimental results show that our descriptor is both efficient in construction and robust to compare with the state-of-the-art methods.