Research On Super-resolution Reconstruction Application Algorithm Of Single Frame Image In Video

Image is the information carrier that can be seen everywhere.In various fields of life,we usually need to use camera equipments to track,collect and shoot videos of selected objects or scenes,and play them clearly on high-resolution imaging devices,and the resolution is one of the important factors to determine the calibration accuracy and quality of the image.However,because of the hardware limitations of the camera and other acquisition instruments and the influence of various conditions in the process of image acquisition,the quality of the original video image is reduced.Image super resolution reconstruction means that we use a certain technological methods to improve the overall quality of the image and people's visual perception of the image,and finally restore the high-resolution image of a target object with more information.Therefore,super-resolution reconstruction has been paid more attention to and increasingly applied to all aspects of society and life,and it has also become the key direction of research.At present,most of the mathematical models of video use image sequences.This paper mainly focuses on the single frame image in the common video sequences in fields of life.In this paper,we first outline the research backgrounds and research status of super-resolution reconstruction of single frame image,the basic principles and the three major categories of research techniques are also introduced.Then,this paper takes the neighborhood embedding algorithm in the super-resolution reconstruction algorithm as the main contents of the research,and mainly carries out two aspects of research:1.This paper improves and optimizes the traditional neighbor embedding algorithm.Through the brief description of manifold learning and neighbor embedding algorithm,this paper proposes an optimized multiple scale iterative neighbor embedding super resolution reconstruction algorithm which using the local geometric similarity between the low resolution patches and the high resolution patches under the small scale factor to reconstruct the image.We collect exemplar high resolution and low resolution patch pairs directly from the given images and their corresponding low resolution images.We use the DCT coefficient of norm luminance as the new extraction feature and reconstruct the image through a process of gradual enlargement.The proposed algorithm can recover more natural looking textures with cleaner and sharper edges in the resulting high resolution images and can get better performance.2.A super resolution reconstruction algorithm based on the optimized image patches is proposed.The algorithm optimizes the image patches of the training sets.We get our training sets by storing the interpolation image patches and corresponding high resolution difference image patches.The DCT coefficient of the image patches is used as a new extraction feature to replace first-order and second-order gradients of the luminance in traditional algorithm.At the end of this paper,an optimized global constraint is proposed to make the reconstructed images more consistent with the global statistical properties of natural images.The algorithm can better protect the local geometry of images and get more details to improve the quality of the reconstructed images.