Video Super-resolution Method Research Based On Similarity Constraints

The number of pixels per unit area in a digital image determines the spatial resolution of theimage and high-resolution image can provide more image details. Super-resolution reconstructionis the process to obtain high-resolution image from one or more low-resolution images with thesame scene. With the development of image processing technology it has wide applicationprospects in the field of medical imaging, video transmission, video monitoring and so on.This article briefly describes the basic theory including imaging theory, observation model,image quality evaluation, and the general methods of super-resolution reconstruction. Then thispaper gives the research about single frame and multi-frames reconstruction, while three singleframe reconstruction algorithms and one multi-frames reconstruction algorithm are provided. Themain work is as follows.First, the algorithm in literature [20] is researched and it is one algorithm that the non-localsimilarity is used to guide the output image reconstructed by the iterative back-projection (IBP)algorithm. The problem of the algorithm is analyzed through experimental and theoretical analysis.In this paper a novel single image reconstruction algorithm using residual post-processing based onstructural similarity constraint is proposed to process the IBP image indirectly and betterreconstruction results to algorithm in literature [20] can be achieved.Second, analyzing the shortcomings of IBP super-resolution reconstruction algorithm, a novelIBP algorithm based on gradient oriented constraints is proposed. Believe that pixels along the edgeof the image have higher similarity. Combined with the gradient oriented and bi-cubic a novelinterpolation based on gradient oriented constraints is introduced to replace the bi-cubic in IBPalgorithm. Then the original isotropic feedback is changed into an adaptive feedback. Theexperimental results show that this algorithm not only can solve the shortcomings in original IBPalgorithm, but also can achieve faster processing speed.Third, the bilateral filtering is studied and the bilateral filtering coefficient templates arereceived from initial interpolation image to correct the error image during the IBP algorithm. Inorder to reduce the complexity of the algorithm edge detecting and edge diffusing operations arepresented, so only the pixels along and across the edges are processed. A novel single-frame IBPalgorithm based on bilateral filtering modification is proposed and the experimental results show that the algorithm can effectively reduce the jagged effect of the image.Fourth, the similarity in video sequences is analyzed and pre-registration is processed toimprove the search accuracy of similarity information. Expanding the residual post-processingmethod described in firs part into the time domain. The operations of bilateral filteringpre-processing, IBP, residual post-processing are repeated twice and then one bilateral filteringpre-processing and one IBP processing are superimposed at last. A novel multi-framereconstruction algorithm based on similarity constraints is proposed and the experimental resultsshow that the reconstructed image has sharp edges and rich texture information.Finally, the work of this paper is summarized, the inadequacies of the research work areillustrated and the direction of future research work is explored.