Obtaining Smooth Directional Field Estimates for Fingerprint Images
Sarat C. Dass, (Michigan State University), email@example.com
Fast and robust estimation of the directional field (DF) is fundamental to the processing of fingerprint images. The estimation of the DF is approached from the point of view of Bayesian statistics. Distributional models are assumed for the observed gradients given the unknown underlying principal gradient directions. Spatial smoothness of the DF in a fingerprint image is modelled using a class of Markov random field priors. The Maximum-A-Posteriori (MAP) estimate of the DF obtained exhibits spatial smoothness while preserving important singularities in the image. We develop algorithms to compute this MAP estimate of the DF in real time. The general framework presented here encompasses previous work on DF estimation as special cases. We also present the results of the DF estimation on each fingerprint image belonging to the Henry class.