Research On Image Super-resolution Reconstruction Based On Image-domain Decomposition

With the development of modern science and technology,image super-resolution reconstruction algorithm has become a research hotspot,and been applied to medical images,military detectives,etc.Image super-resolution reconstruction is trying to figure out how to reconstruct a high-resolution image from a low-resolution image.This paper puts forward two effective modified methods and an application scene based on image-domain decomposition method and image super-resolution reconstruction method proposed by Yang.One is an image super-resolution reconstruction algorithm based on OSV model as well as its Split Bregman method and sparse representation.The other is an image super-resolution reconstruction algorithm based on incorporating LBP and GLCM and sparse representation.Another is an application scene based on the combination of salient map and image super-resolution reconstruction based on sparse representation.The super-resolution reconstruction algorithm based on OSV model as well as its Split Bregman method and sparse representation uses image decomposition method based on OSV model and its Split Bregman to decompose the image into structure part and texture part.We can take different methods to reconstruct structure part and texture part.High resolution image can be rendered out with the combination of these two parts.The super-resolution reconstruction algorithm is built on the premise of image super-resolution reconstruct method proposed by Yang,while incorporating both LBP and GLCM methods in order to categorize image patches into structure patches and texture patches.We can train different patches into different types of dictionary pairs which would be utilized correspondingly to reconstruct image patches,and the high resolution image can be rendered out with the combination of different patches.The super-resolution reconstruction algorithm based on saliency map and sparse representation is incorporated with salient region detection theories as well as FT algorithm.With the complementary implementation of binarization of salient map achieved by FT algorithm,defining whether it's a salient area or not based on the existence of salient points,utilizing Yang's algorithm on the salient areas and bicubic interpolation on non-salient ones,we would be able to reach the desired effect.Experiment results demonstrate that the first two methods are better than Yang's algorithm in the way that the reconstructed image is smoother and clearer.At the same time,the third method greatly reduces the time that Yang's algorithm requires without losing the subjective quality of the image.