Finger-Print Image Super-Resolution via Gradient Operator based Clustered Coupled Sparse Dictionaries

Abstract

In this paper, a novel approach is employed for fingerprint image super-resolution based on sparse representation over a set of coupled low and high-resolution dictionary pairs. The primary step of fingerprint super-resolution involves learning a pair of coupled low- and high-resolution sub-dictionaries for each cluster of patches sampled from training set of fingerprint images. The clusters are formulated based on patch sharpness and the dominant phase angle via the magnitude and phase of the gradient operator for each image patch. In the reconstruction stage, for the low-resolution patch the most appropriate dictionary pair is selected, and the sparse coding coefficients are calculated with respect to the low-resolution dictionary. The equality assumption of the sparse representation of the low and high-resolution patches is the link between the low and high-resolution features space. For the reconstruction of high resolution patch, the sparse coefficients calculated for low-resolution patch are directly multiplied with corresponding high-resolution dictionary. The conducted experiments over fingerprint images show that the algorithm is competitive with the state-of-art super-resolution algorithms.