A Novel Super-resolution Approach Based on Supervised Canonical Correlation Analysis

Abstract

-In this paper, we use supervised canonical correlation analysis (SCCA) method to extract features which maximize the correlation between HR and LR face images. Then Relationship Learning (RL) is used to construct the mapping relationship between the face coherent features. SCCA method comprehensively considers the within-class information and the similarity of HR and LR images, to make the SR image closer to original HR image. Experiments on Yale and ORL face databases show that our method has higher recognition rate.