Channel refinement of fingerprint pre-processing models

Deep models are the state-of-the-art models for fingerprint pre-processing. However, these models have very high number of parameters, usually in millions. As a result, redundancy is observed among the features learnt by deep learning-based fingerprint pre-processing models. A popular technique to help deep models learn distinct and informative features is channel refinement. A recent study has illustrated the capability of channel refinement to improve generalization of fingerprint enhancement models. Motivated by the above-mentioned study, this chapter delves into presenting a detailed study illustrating the usefulness of channel refinement in reducing redundancy and imparting generalization ability to fingerprint enhancement models. Furthermore, we extend this study to assess whether channel refinement generalizes on fingerprint region of interest (ROI) segmentation. Extensive experiments on 14 challenging publicly available fingerprint databases and a private database of fingerprints of the rural Indian population are conducted to assess the potential of channel refinement on fingerprint pre-processing models.