Learning Multiple Linear Mappings for Efficient Single Image Super-Resolution

摘要: Example learning-based superresolution (SR) algorithms show promise for restoring a high-resolution (HR) image from single low-resolution (LR) input. The most popular approaches, however, are either time- or space-intensive, which limits their practical applications in many resource-limited settings. In this paper, we propose novel computationally efficient SR method that learns multiple linear mappings (MLM) to directly transform LR feature subspaces into HR subspaces. particular, first partition the large nonlinear space of images cluster Multiple subdictionaries then learned, followed by inferring corresponding based on assumption LR–HR features share same representation coefficients. We establish MLM input desired outputs order achieve fast yet stable recovery. Furthermore, suppress displeasing artifacts generated MLM-based method, apply nonlocal means algorithm construct simple effective similarity-based regularization term enhancement. Experimental results indicate our approach is both quantitatively and qualitatively superior other application-oriented methods, while maintaining relatively low time complexity.