Image Super Resolution Reconstruction Based On Shearlet Fusion

Super-resolution image reconstruction algorithm is one of the topics at the forefront of today’s field of image processing, it has important meanings in improving the image quality. The core idea is how to use signal processing method to produce single high-quality, high-resolution images from one or multiple images with complementary information. The technique requires no hardware support, image resolution can be improved, and thus low cost, and can achieve effective utilization of resources acquired image.Shift-invariant shearlet transform is a new multi-scale analysis image processing tool, with a good translation invariance to overcome the traditional wavelet translation invariant shortcomings. It also can be localized to analysis of the image signal, as in image analysis provides more abundant phase information. This thesis mainly based on shift-invariant shearlet fusion algorithm of super resolution reconstruction. The main work is as follows:1. First of all,it summarizes current research image super-resolution algorithm, analyzed the image degradation model, the basic knowledge of super-resolution reconstruction and performance evaluation. A brief summary of the basic theory of wavelet analysis is summarized, focusing on the shearlet transformation and shift-invariant shearlet transform’s concepts.2. For a single image super-resolution (SR) reconstruction problem, a novel image SR method based on SAI-Bicubic interpolation and shift-invariant shearlet transform fusion is proposed. Firstly, the source image are separately interpolated by Soft-decision adaptive interpolation (SAI) and Bicubic interpolation, then the shift-invariant shearlet transform (SIST) is adopted to decompose the two interpolated images in different scales and directions, and the low-frequency and high-frequency sub-band coefficients of the two images are obtained.For the low frequency sub-band coefficients, according to the regional variance to determine the fuzzy similarity, a adaptive weighted fusion rule combined with improved sigmoid function is presented; while for the high frequency sub-band coefficients, using a new Sum-modified Laplacian (NSML) and combine with the weighted average fusion rule. Finally, the high resolution image is obtained by performing the inverse SIST on the combined coefficients.Experimental results indicate that compared with other reconstruction algorithms, the proposed SR method can reconstruct higher quality results both quantitatively and perceptually.