Image Super-resolution Recovery Of The Variational Pde Method

Super-resolution (SR) restoration is one of the most important research subjects inthe field image processing and it becomes more and more attractive in this field duringrecent years. It has a researching significance in theoretically as well as a broadapplication in practically.Super-resolution restoration is an economic image processing method thatextracting higher resolution images containing more details from an image sequence oflower resolution, through such methods as motion estimation, image fusion, deburingand denoising.This dissertation focuses on the Super-resolution restoration frame based on thebilateral filtering.The primary reaserching content includes assessment of theperformance of SR restoration, Research of the bilateral filtering based SR restorationmodel and its algorithm frame, and the potential function and the similarity functionunder the model.According to the criteria of similarity and quality of images, we conduct theresearch in both visual quality and subjective measure aspects, and then propose anassessment system of image quality, which is the basic criterion of the algorithm andmodel of image SR restoration.In the aspect of image SR restoration, from the statistical perspective, we could useBayesian and Markov random field theory to add a prior probability model of image asthe regularized item, and then develop a PDE evolving based iterating algorithm frame.For the framework discussed above, to choose both a suitable grayscale variationrelated potential function and a pixel Euclidean distance related affinity function has agreat impact for the SR reconstruction. We compare a few effective potential functionssuch as Tikhonov, TV, Huber and entropy function in the SR restoration framework.Furthermore, we propose a differentiable hybrid function between the negentropyvariational integral and total variation function. Experiment results demonstrate theeffectiveness of our approach, both in the visual effect and the other measure values. Onthe other hand, we propose a general affinity function in the SR restoration framework.This function reflects the similarity of eculidean distance and features. Experimentalresults prove that the proposed frame is effective for images with different textures anddetails.