Improving feature based dorsal hand vein recognition through Random Keypoint Generation and fine-grained matching

Recently, SIFT-like approaches have shown their advantages of performance and robustness in dorsal hand vein recognition. This paper presents a novel method to recognize the vein pattern of the dorsal hand, which discusses two important issues in the SIFT-like framework, i.e. keypoint detection and matching. For the former, a Gaussian Distribution based Random Keypoint Generation method (GDRKG) is proposed to localize a sufficient set of distinctive keypoints, which largely reduces the computational complexity of the state of the art ones, such as DoG, Harris, and Hessian. For the latter, a Multi-task Sparse Representation Classifier (MtSRC) based fine-grained matching strategy is introduced instead of traditional coarse-grained matching, to precisely measure the similarity between the feature sets of the samples. The proposed method is tested on a dataset of 2040 vein images of 204 dorsal hands, and it outperforms the state of the arts clearly proving its effectiveness.