Abstract: Aiming at the problem that maximum margin correlation filter (MMCF) does not consider the internal structure information of the samples and causes insufficient training, minimum class locality preserving variance correlation filter (MCLPVCF) was proposed. MCLPVCF fuses the idea of sample weighted adjacency graph, and introduces the locality preserving within-class scatter, and considers distribution information and the intrinsic manifold structure of the samples. Meanwhile, MCLPVCF maximizes the classification margin and optimizes the correlation output, and then takes into account the category information of the sample during the training process. The experimental results show that compared with MMCF and other traditional correlation filters, the proposed method has a great improvement in performance.