The Application Of Perceptual Compressive Sensing In Face Super Resolution

As the development of information age, the acquisition andprocessing of information become more and more important. As for image,the resolution shows the amount of information contained in it. Superresolution (SR) is a way to reconstruct high resolution (HR) image fromlow resolution (LR) image. How to restore the HR image with higherquality using the limited information from the LR image is the mostimportant research topic in SR field.This paper summarizes the existing SR algorithms, and researches theSR method based on sparse representation in detail on the basis of thatsummary. The method learns the over complete dictionaries of both HRand LR training images with common sparse representation, and appliesthem to do the super resolution reconstruction for testing LR image. Thispaper wants to apply the human vision system (HVS) to this method, inorder to make the SR image have more sensitive information for eyes.Therefore, we first promote the perceptual compressive sensing methodbased on texture model in HVS, and prove it work well in thereconstruction effect through experiments. After that, we apply the methodinto the construction of feature vector and regularization parameter in SR.According to our simulation experiments, although the algorithm of featurevector does not work very well, it proves the existing method is reliable;while the algorithm of regularization parameter can make the SR imagehave better effect both in vision and in objective evaluation system, almostwithout increase of computational complexity.The research in this paper of SR based on perceptual CS also proves that HVS can play important roles in vision processing field such as imageprocessing system and worths more research in the future.