Research Of Super Resolution Reconstruction Technology Based On Image And Video

With the rapid development of digital multimedia content such as images andvideos, the requirement for the quality of images and videos is much higher andhigher. In response to the growing social demand, it is urgent to develop a technologycapable of improving the spatial resolution. Presently, super resolution reconstructionis an active topic in the field of image processing for researchers. It is a softwaretechnology which can overcome the existing level of an imaging system and improveresolution without changing the hardware equipment. Because of the economy andpracticability for the super resolution, it has got the favor of many researchers, beenwidely applied in many fields such as video surveillance, battlefield monitoring,medical diagnostics, satellite remote sensing, high sensitivity digital television, etc.Super resolution reconstruction is a subject with great theoretical significance andpractical value.After a brief introduction of the theory for super resolution reconstruction, thispaper carries on a classification and summary for the existing techniques. Researchhas been done on super resolution reconstruction aiming at improving the spatialresolution. The main works are summarized as follows:Firstly, for the process of traditional IBP super resolution reconstruction, fails toconsider the structure feature of images, causing the details of the reconstructedimaging edge are not ideal, a novel method is proposed. Through the pre processingby interpolation technology based on structure tensor, to improve the quality of initialestimated image, so as to increase the convergence speed of reconstruction results,and a bilateral filter is introduced to revise the extracted edge of image, to retain andenhance the edge information of it. Experimental results show that the improvediterative back projection method improves the visual quality of reconstructed image,protects the edge information. Compared to the traditional IBP super resolutionmethod, blurs and artifacts in the edge are suppressed effectively.Secondly, for the computational complexity problem of accurate motionestimation between sequential images, a method based on edge detection operator andstructure tensor is proposed. This method is based on the in-depth study of theprinciple about kernel regression, introducing a3D Sobel operator to revise thegradient estimation of steering kernel function; according to the local estimation of structure tensor for each pixel neighborhood in video frames, obtain3D localstructure informaion of interested pixel point, average weight the neighboring pixelscan achieve estimation of targeted position; And a geometric distance function isfused into the kernel, modifies the pixel deviation and empty phenomenon of steeringkernel regression. Experimental results show that the improved method significantlyimproves the reconstruction quality, avoids the problem of accurate motion estimation,and generates images with higher subjective and objective quality.At present, the super resolution reconstruction research is still a hot topic. Thereare many related literatures. However, this study can promote the further developmentof super resolution techniques, enlarging the application range of this technology.Aiming at two kinds of super resolution reconstruction methods research, preliminaryresults are achieved.