Sparse Representation for Image Super-Resolution

摘要: This chapter concentrates the problem of recovery a high-resolution (HR) image from single low-resolution input image. Recent research proposed to deal with super-resolution sparse coding, which is based on well reconstruction any local patch by linear combination an appropriately chosen over-complete dictionary. Therein LR (Low-resolution) and HR (High-resolution) dictionaries have be exactly corresponding for reconstructing patterns. However, conventional coding usually achieves global dictionary D=[D l ; D h ] jointly training concatenated patches, then reconstruct as separated . strategy only can achieve minimum error cannot obtain dictionaries. In addition, accurate coefficients using are also unable estimated Therefore, this paper proposes firstly learn features propagates one, called HR2LR propagation, mathematical proving statistical analysis. The effectiveness propagation in demonstrated comparison approaches such interpolation.