Semi-Coupled Dictionary Learning Based on Sharpness Measure for Single Image Super Resolution

Abstract

In this research work, a new algorithm is simulated based on semi coupled dictionary learning (SCDL) for single image super resolution (SISR) problem. Semi coupled dictionaries were planned for set of clustered information. The clustered information classified in three clusters by sharpness measure based and included only those patches whose sharpness measure value is same. For task of super resolution SR, invariance of sparse representation assumed. Formerly mapping function with a pair of dictionaries were initialized for each cluster, Afterword the dictionaries, mapping function and sparse coefficients were updated. During the reconstruction phase the dictionary pair and mapping function of respectively cluster are used to recover HR image. The LR and HR dictionary pair with mapping function are selected which give the slightest sparse representation error. For high resolution patch approximation, dictionaries pair with mapping function of that cluster are utilized. In addition, it’s also tried modifying the results, by observing power spectral density (PSD) of distinct images through computing sharpness-based scale-invariance ratio for patches that categorized in three clusters. The proposed work is compared with the previous research of image SR algorithms. By proposed procedure the recovery of HR image feature becomes prominent.