Abstract:The traditional fingerprint image recognition method can not solve the problems of large amount of data, high sample dimension and non-linear sample data. Therefore, the dynamic sparse representation method is applied to the non-contact fingerprint image recognition. The problem of color deviation, partial occlusion and content deletion in the collected samples is solved by the low-rank matrix recovery. By dividing the image to be subdivided into several sub-image blocks by the method of local area division, the dimension conversion is completed. This paper introduces the criterion that can measure the local structure of the sample to be identified and the training sample, and improves the objective function by combining the dynamic active set, and establishes the dynamic sparse representation model. The sparse representation coefficients of each sub-image are solved, all the reconstructed errors of the non-contact fingerprint sub-images are obtained according to the sparse representation coefficients, all the errors are merged, and the category with the smallest error is assigned to the fingerprint image to be identified to realize the non-contact fingerprint image Recognize. Experimental results show that the proposed method is practicable and more reliable than other methods.