Study On The Hybrid Multi-frame Image Super-resolution Reconstruction Methods

In recent years, with the introduction of sparse representation theory in the field of image super-resolution, there emerging many new algorithms promotes the development of image super-resolution reconstruction technique. Now, more and more researchers concern and pursue more information from the low resolution image, in order to improve the quality of image super-resolution reconstruction. Currently, the technology of super-resolution image reconstruction has been widely used in remote sensing, mapping, military detection, security surveillance and medical imaging and other areas.Based on the technology of common single and multi-frame image super-resolution reconstruction, this article propose a new hybrid model of multi-frame super-resolution reconstruction to providing higher multi-frame reconstruction resolution. The whole work for this article mainly contains the following sections:1. Learn and achieve several common image super-resolution reconstruction methods in depth, such as those methods based on interpolation, learning and multi-frame reconstruction.2. Based on the sparse representation sparse dictionary coding, image registration, morphological component analysis (MCA) model, and this study give an implementation method of single frame image super-resolution reconstruction by MCA and residuals compensation. Firstly, use MCA method to decompose the low-resolution image, extract its cartoons component and texture component. Secondly, use sparse K-SVD dictionary learning methods respectively training, to generate the corresponding cartoons and texture sparse dictionary. Thirdly, calculate the MCA decomposition residual and determine the component of residual compensation.Finally, respectively use cartoon reconstruction, texture reconstruction with the component of residual compensation to rebuild the image. The experimental results show the way to achieve super-resolution image reconstruction is very effective, and can preferably recover the details of the image.3. In the study, based on the research method of the multi-frame image registration, propose a new method that combines Binary Robust Invariant Scalable Key points (BRISK) feature extraction with the Projection Onto Convex Sets (POCS). The method adopt BRISK feature extraction algorithm to enhance precision of multi-frame image registration, improve image super-resolution reconstruction accuracy. Experimental results show that the proposed method outperforms traditional POCS reconstruction methods and have better quality of reconstruction image.4. For the traditional frame image super-resolution reconstruction method in the larger reconstruction multiples lead to unsatisfactory quality. In this paper, we propose a new method for image super-resolution reconstruction based on the hybrid model, which is using POCS combined MCA and residual compensation. This method is based on the combination of method base on multiframe reconstruction and method base on learning. Experiments show that this method can achieve high resolution image reconstruction.