Bank customer identification has a far-reaching significance, which can identify the bank's important customers by using face recognition algorithm in different application scenarios. When the feature information of face is changed or masked, the accuracy of recognition will be greatly reduced. The paper introduces the face layered filtering algorithm to remove non characteristic information (such as beard, glasses, scars, etc.) before doing face recognition, so that the face recognition target has a higher unity. By using MUCI data sets to verify the feasibility of facial layered filterer, we add the face layered filtering algorithm in unified recognition module to improve the recognition efficiency. And the feasibility of the algorithm is verified.