The Study Of Sequence Images Super Resolution Technology

Image has some features of clear and vivid with intuitive, has been used as a way forstorage and share visual information. With the rapid development of science andtechnology and the improvement of human living standards, people’s requirements on theimage quality are also increasing, but image quality is determined by image resolution, it isa measure of the performance capabilities of image detail. In the same scene, the higher theresolution of the image, the more detailed of the scene details information, the intuitivepoint of view seems more clear, otherwise, the image will blurred due to less details. Sopeople want to get high-resolution images directly, but it need to get a higher resolutionimage through hardware imaging device, which is more stringent requirements forhardware equipment, and increased production costs, in order to solve this problem, thesuper-resolution reconstruction is raised in response to the proper time and conditions.Super-resolution image reconstruction is digital image processing technology, whichuse a frame or more low-resolution images to reconstruct an image with high resolution.Super-resolution reconstruction is the inverse problem of image degradation, but becausethe result of the reconstruction is not unique, so it is a typical ill-posed problem. Hence itneed to use a priori constraint to regularize the problem, so the use of regularizationalgorithm to solve these problems. From the beginning of the total variation regularizationalgorithm, and later the bilateral filtering total variation regularization algorithm, thereconstruct effect is getting better and better. Because the image acquisition process will beaffected by the moving of object scene or cameras, atmospheric disturbances and sensoroptical system itself, making the image blur or noise lead to degradation of image qualityand low image resolution. The main types of noise existed in the image is Gaussian noiseand impulse noise (salt and pepper noise).Regularization algorithm contains two functions, the first term is the fidelity item, itsaim is ensure the closeness of the reconstructed high-resolution images and the original sequence low-resolution images, the higher the closeness level, the more degradationfactors through analyze in line with the actual situation, there are L1and L2two norms, useL1norm can filter out image impulse Laplace noise as well as protect image edge, to filterout Gaussian noise, L2norm can also smooth image; The second item is regular item, inorder to solve the ill-posed problem of super resolution reconstruction and make sure thereconstruction results unique, it contains the regularization parameter, which is to balancefidelity item and regular item.Simultaneously filtered Gaussian noise and impulse noise and reconstruct highresolution images, this paper presents a paradigm based on L1and L2mixing andcombining bilateral total variation (Bilateral Total Variation, BTV) regularizedsuper-resolution reconstruction of image sequences. First of optical flow field modelstrategy based on multi-resolution low-resolution image sequence registration to achievesub-pixel accuracy; secondly full advantage of L1and L2hybrid paradigm, with BTVregularization algorithm to solve the super-resolution reconstruction the morbid inverseproblem; and finally use the proposed algorithm for sequence image super-resolutionreconstruction. Experimental data show that the proposed algorithm can reduce the imageof the mean square error and improve the peak signal to noise ratio. Experimental resultsshow that the proposed algorithm can effectively filter Gaussian noise and impulse noise,keep the image edges to improve the image recognizable, can provide a good technicalfoundation for the license plate recognition, face recognition and video surveillance andother aspects.