Research Of Video Compression Based On Super-Resolution

This thesis focus on video compression based on super-resolution. Every user of multimedia technology expects good image and video quality from the receiver. Unfortunately, due to factors like channel capabilities's limitation, video sequences are often down-sampled, transmitted or stored at low bitrates, resulting in a degradation of their final visual quality.In this paper, a framwork is proposed for intentionally introducing down-sampled high resolution (HR) image sequences before compression, and then utilizing super resolution (SR) techniques for generating an HR video sequence at the decoder. Segmentation is performed at the encoder on an image or group of images, which is used to guide the down-sampling process. It is also encoded and sent to the decoder as side information in order to guide the SR process.Some algorithms about down-sampling and compressed image super-resolution are proposed in this paper. (1) Down-sampling. The method of region segmentation is used for guiding the process of down-sampling high resolution (HR) image sequences. Firstly, each frame is segmented into three kinds of blocks(motion blocks, no motion and flat blocks, no motion and textured blocks), which used a kind of motion measurement(MM) to segment the motion blocks and no motion blocks, and used canny operator to segment flat blocks and textured blocks. Secondly, there are three down-sampling means according to the different segmenting regions. (2) Compressed image super resolution.Different compressed image super-resolution methods are exploited accordin-g to different segmented regions, which used interpolation technology for flat blocks, used super-resolution technology based on image registration for motion blocks, and used AMR technology for textured blocks. The experimental results validate the usefulness of the proposed method, improving compression ratio efficiently on the premise of high decoding frame quality.