Face image super-resolution using 2D CCA

摘要: In this paper a face super-resolution method using two-dimensional canonical correlation analysis (2D CCA) is presented. A detail compensation step followed to add high-frequency components the reconstructed high-resolution face. Unlike most of previous researches on algorithms that first transform images into vectors, in our approach relationship between and low-resolution image are maintained their original 2D representation. addition, rather than approximating entire face, different parts super-resolved separately better preserve local structure. The proposed compared with various state-of-the-art multiple evaluation criteria including recognition performance. Results publicly available datasets show super-resolves high quality which very close ground-truth performance gain not dataset dependent. efficient both training testing phases other approaches. HighlightsA new (SR) CCA presented.The works directly without reshaping vector.A further enhances images.Experimental results outperforms current SR methods.The computationally due small matrices involved.