Performance Improvement in a Fingerprint Classification System Using Anisotropic Diffusion

In a previous work, [1], we evaluated a classification algorithm based on the Karu-Jain method [2] and compared the performance with a fully manual method used at the Direccion Nacional de Identificacion Civil (DNIC). In this paper, we analyze the high performance improvement achieved using anisotropic diffusion instead of pure averaging for the directions smoothing. We also define a quality measure that shows high correlation with the experts’ criteria. The results are evaluated over 2800 images extracted from a 4 million fingerprint card archive maintained by DNIC.