Based On The Study Of Video Super-resolution Research And Application

In recent years, the field of video surveillance has been actively developed. Video surveillance system has developed from analog video surveillance system to digital video surveillance system, from analog low-resolution cameras to high resolution digital cameras. Monitoring equipments have developed from ordinary monitor developments to professional monitors, and even have dedicated TV walls. Image/video technology in these areas has been increasingly widely used.The real-world scenes contain a wealth of information, especially the high-frequency data. Because of the limitations of the imaging conditions and the use of the camera image acquisition system for digital image, it is difficult to obtain all the information in the actual scene. Front-end collection of the resolution limit of the camera sensor, results in the loss of high frequency image information. With the development of digital video surveillance systems, the video surveillance system needs to display high-resolution high-definition video images. Therefore, the super-resolution algorithm has been fully developed. Super-resolution algorithm is a method that uses one or several low-resolution images images to obtain the high-frequency information of the picture, and makes a super-resolution image.The traditional image amplification algorithm is based on the interpolation method. It has high computation speed, but is not able to use a priori knowledge of the image, which will be easy to cause a jagged effect. Reconstruction-based super-resolution algorithm can dig out the actual low-resolution images obtained in the high-frequency information and complementary information between multiple images, combined with a priori knowledge of the image, to form the high-resolution images. But the large amount of calculation complexity is high.Junping in learning-based super-resolution algorithm, the neighborhood embedded in the case of small training sample, the use of edge detection and feature selection methods, to obtain high-resolution images from low resolution images. Type used in the algorithm of image edge features to speed up the search process of the training images. The rotation of the training sample has been more high-frequency information, to make up for the few shortcomings of the training samples. Through the experimental comparison of learning-based super-resolution algorithm can get a better image than the traditional interpolation methods. In this paper, this super-resolution algorithm, an analysis of the total number of training samples, the impact of the training sample subset obtained video surveillance of the best design programs.In the actual use of video surveillance, the artical uses the learning-based super-resolution algorithm to design a network video decoding system. This system can satisfy the existing video decoding system based on function, and the low resolution images can be amplified to high resolution image using hardware super resolution function.