A Fast and Stable Bold Feature Extraction Algorithm

Abstract

To improve the stability and real-time of bold feature extraction, this paper proposes a fast and stable bold extraction algorithm (FSB). In the feature detection stage, three-layer Laplace filter is used to construct image scale space instead of binary filter, and a local feature detector is built. In the stage of the feature description, the binarized difference of integral image is used as contextual information to build local feature descriptor. The experimental results show that compared with SIFT and SURF, the speed of FSB is highly improved while their performances are similar. It is suitable for real-time application system.