Multi-scale dictionaries based fingerprint orientation field estimation

Orientation field estimation is significantly important for fingerprint recognition. Dictionary based algorithm and its variant, localized dictionaries based algorithm have shown promising performance. In this paper, we extend the original dictionary based algorithm to a multi-scale version. The motivation is that small scale dictionary is more accurate while large scale dictionary is more robust against image noise. Hence information from orientation fields of different scales can be integrated to obtain better results. A multi-layer MRF model is used to formulate and solve the proposed problem. Experimental results on challenging latent fingerprint database demonstrate the advantages of the proposed algorithm.