Abstract: Aiming at the problems of strong local pixel similarity and poor recognition effect in a single direction of finger vein images collected under near-infrared light, a finger vein recognition method based on template voting and multi-directional fusion was proposed. Firstly, the template voting local three pattern (TVTP) was proposed to fully use the information of local multi-neighborhood points and reduce local pixel similarity. Secondly, based on the rich directional feature information contained in finger vein images, the multi-directional coding (MDC) was proposed to acquire directional features with image discernment, strengthen the robustness between different directional features, and solve the problem of poor recognition rate of a single direction. Finally, the block histogram was used for statistical features and collaborative representation (CR) was used to improve the recognition efficiency. The experimental results show that the recognition rate of the proposed method on the SDUMLA dataset, USM dataset, and THU-FVFDT2 dataset reaches 99.32%, 99.73%, and 99.75%, respectively. Compared with other classical and novel algorithms, the proposed method achieves better recognition effects, meets real-time requirements, and has application values.