Research On Video Super-resolution Technology

As the rapid progress and popularization of the hand-held photographic equipment, it is more convenient to acquire the pictures and videos. However, the users for the demand of the picture quality keep increasing. To solve the problem of clear display for the low resolution video playing on the high resolution screens, signal processing method should be adopted to enhance the spatial resolution of video. Video super resolution is widely used in Internet videos, digital videos, public safety and so on.This paper firstly introduces the background of super resolution and research meaning. Secondly, the induction of research current situation, such as multi-frame super resolution, video super resolution and single-frame super-resolution method is presented. Then the imaging model of low resolution videos and the reconstruction of high resolution are introduced. According to the type and amount of the reference frames, two aspects are researched, including intra-frame redundance single-frame super resolution and inter-frame redundance video super resolution.Improved self-similarity characteristic based single-frame codebook super resolution is proposed. Firstly, self-codebook generation is the same as the codebook algorithm. The image downsampled by two times from current low resolution image is processed and added to the self-codebook. Then several similar low-blocks which are conforming to the match principles are searched. Obtain the weight according to the similar degree, sum the corresponding high-frequency blocks and finally get the high-frequency block. Then, add the high-frequency component to corresponding linear interpolation image and get the initial estimation image. The image is finally adjusted by luminance-based variation regularization. Experimental results show that improved algorithm can gain details and reduce the search domain. Regularization processing method makes edges distinct.A video super resolution algorithm is proposed combining gradient regularization with motion estimation based on inter-frame prediction and delamination technology. Three pyramid delamination images are gained through up-sampling and down-sampling by two times. Take the upper motion vectors as the motion estimation starting point of the next layer. Adaptive threshold block matching is adopted on every layer. Then error motion vectors correction is done according to the direction difference and amplitude difference with the arrounding motion vectors. Motion compensation patches are obtained through the motion vectors and using these blocks to reconstruct the image. Gradient variation regularization is adopted to reconstruct the videos. Experimental results show that proposed video super-resolution algorithm can enhance the spatial resolution and reduce the edge surge effectively.