Double Layer Super-resolution Recon-struction Based On Promoted Sparse Coding

High resolution(HR) image is of great significance in the practical application ofmedical diagnosis, satellite monitoring, security monitoring etc.. However, the way byimproving the imaging system hardware to improve the image resolution is high cost,and it needs to solve some technical problems in the process of imaging. Imagesuper-resolution(SR) is mainly through various computer algorithms, using observedlow resolution(LR) image to reconstruct high resolution(HR) image. It solves theproblem of high cost and complex process which caused by improving hardware toobserve HR image, and it causes widespread concern in the field of image processingwith its high quality reconstruction results and low cost.In this paper, we research on the natural image super-resolution reconstructiontechnology, through in-depth analysis of the super-resolution reconstruction algorithmbased on sparse representation, a double layer super-resolution reconstruction algorithmbased on promoted sparse coding is proposed, and related experimental demonstrationcontrast are executed. Finally, in order to further improve the quality of reconstructedimage, we improve the post-processing model in the reconstruction algorithm via sparserepresentation, and put forward a kind of global-local constraint post-processing model.The main research contents of this paper are as follows:(1) This paper introduces the super-resolution technology background, basicconcept, significance and the prospect of its application, deeply analyzes of the imagedegradation process, and this process is described by the mathematical method ofanalysis, and establishes the model of image degradation. Then we analyze the currentdevelopment status of image super-resolution technology at home and abroad, introduceand summary the classic and representative algorithms in three categories ofsuper-resolution reconstruction algorithm, introduce the difficulties of the research, andlay the foundation for later research. Finally, the evaluation indexes of super-resolutionreconstruction of image are descried and analysized.(2) This paper explains the principle of sparse representation of signal and imagein detail, introduces three sparse representation methods in compressive sensing, andone of them which based on over-complete dictionary is introduced detaily. Then weintroduce the principle of reconstruction algorithm model based on the sparse representation, including the constraints of sparse representation reconstruction,reconstruction algorithm and dictionary training algorithm.(3) This paper research the super-resolution reconstruction algorithm based onsparse representation deeply, and on this basis, we propose a double layersuper-resolution reconstruction framework based on promoted sparse coding. In theproposed algorithm, according to the existing sparse coding methods for single-imageeasily lead to incorrect geometrical structures in reconstructed images, we propose asparse coding method combining the incoherence constraint of dictionary and thenonlocal self-similarity constraint of sparse coefficient. Then, to solve the problemof lossing of high frequency information of the reconstructed images because ofintroducing the nonlocal self-similarity constraint, such as the phenomenon ofover-smoothing and fuzzy in the regions of steep gradients such as edge, we alsopropose a double layer reconstruction scheme based smooth layer (SL) and texture layer(TL), the HR texture image (HRTI) is recovered using the promoted sparse codingmethod, the high resolution smooth image (HRSI) is reconstructed using a globalnon-zero gradient constraint SR model, then two layers of the reconstructed image aremerged into the initial high resolution image. The experimental results show that thealgorithm proposed in this paper is effective.(4) By analyzing the reasons for the phenomenons that the possible edge blur andflaw in reconstruction image, we put forward a global-local constraint post-processingmodel. At the same time ensuring the global reconstruction constraints, utilizing theself-similarity to local structure and the robustness to noise of the steering kernelregression to deblur, denoise and eliminate some fuzzy of the reconstructed image, and,enhance the robustness of the algorithm. The experiment proved that the modeleffectively improves the image quality.