Fingerprint orientation modeling by sparse coding

Local ridge orientation field describes well the topological pattern of fingerprint ridge-valley flows. It is a rich information resource for fingerprint image processing and feature extraction in automatic fingerprint recognition algorithm. But reliable estimation of orientation field is still challenging for fingerprint images of poor quality. In this paper, we propose a method for modeling fingerprint orientation field using sparse coding. The basis functions of discrete cosine transform (DCT) are used to build the basis atoms for the representation of orientation field and l1-norm regularized optimization is used for the sparse coding of DCT atoms. Finally, orientation field is reconstructed by linear combination of sparse coefficients and DCT atoms. The proposed orientation model does not need any prior information such as the locations of singular points and it is easy to implement. More importantly, the effect of noise can be significantly reduced by sparse coding. Experimental results and comparison are presented to show the effectiveness of the proposed method for modeling orientation fields of fingerprints, especially the poor quality fingerprints.