Research For Color Image Super Resolution Reconstruction Algorithm Based On Sparse Representation

With the improvement of people's living standards and the needs of image processing applications, the resolution of image is required higher and higher. But it is always with high cost and big risk improving the resolution of image through hardware. So it is hard to be extended in real application. Image super-resolution is the technology to obtain higher resolution image based on signal processing and becomes a rapidly growing field in image processing. It takes a single low resolution image or multi low resolution images as input, uses signal processing technology to generate a higher resolution image. It plays an important role in video, remote sensing, medical and military field. Recently, learning-based super-resolution becomes a hot topic in this field.This dissertation reviews the super-resolution related theory and classical algorithms systematically, and emphatically introduces the algorithm of the super-resolution based on sparse representation. Generally, the processing objects of the super-resolution algorithms are gray images. Considering the difference between color image and gray image, the dissertation mainly studies on super-resolution algorithms of color image based on sparse representation. In order to reduce the training time of dictionary, introduce a fast and effective method of the structure of dictionary. In the past, the regularization parameter of iterative algorithm always choose constant, but we propose adaptive regularization parameter in this dissertation and realize image super-resolution reconstruction adaptively. No matter the image has simple texture or complex texture, the proposed algorithm gets good results and shows good performance. In the past, when dealing with noise image, the traditional super-resolution algorithms always divide this problem into two steps:first, image de-noising; second, super-resolution reconstruction. The dissertation adds the noise model into the super-resolution algorithm, and makes de-noising and super-resolution reconstruction work simultaneously.The results shows that the method proposed in this dissertation can reconstruct high resolution image. The reconstructed image is improved significantly, no matter on subjective visual or peak signal to noise ratio.