Image Super-Resolution Using Compressed Sensing Based on Learning Sub Dictionary

Many applications have benefited remarkably from high resolution (HR) imaging models such as, astronomical and biomedical imaging system. In recent years, image enhancement and resolution approach has been embark the great result, and received great deal of attention of the researcher, and many researchers have proposed several methods to achieve the goal of high resolution image. In this paper an efficient method of high resolution image based on the concept of Compressed Sensing (CS) have been introduced, which uses sub dictionary instead of redundant dictionary and traditional orthogonal basis. The new framework is consisted of three phases. Firstly, we designed the sub dictionary that are learned from a range of datasets of high quality patches and then selected adaptively. Secondly, Principal Component Analysis (PCA) has been applied to each data sets of the high quality patches to evaluate the principal component from which the dictionary is constructed. Finally, the HR image is generated by averaging all high resolution patches. In addition, the proposed method has been demonstrated better results on real images in terms of peak to signal and noise ratio (PSNR), structural similarity (SSIM) and root mean square error (RMSE). Furthermore, our method has been evaluated by deriving the modulation transfer function (MTF), the MTF curve showed better reconstruction of HR image and achieved various improvements compared with other methods. Key-Words: Compressed Sensing, Compact Dictionary, Image Resolution, Modulation Transfer Function, Principal Component Analysis (PCA)