Research On Enhanced Prediction And Postprocessing Of Video Coding

In the 21st century, the pace of development of high technology speeds up. Video phone, video conference, IPTV and HDTV etc are very popular among everyone's life. All these applications result in huge number of video data, thus become a big challenge to video storage and video transmission. One of the most important solutions to this challenge is video coding, also named as video compression. The main objective of video coding is to reduce the amount of video data (number of bits) needed to represent or describe the video while preserving the video quality. Through research of about six decades, video coding has been formed as the hybrid coding framework consisiting of three core components: predictive coding, transform quantization coding and entropy coding.As a core technology, predictive coding has contributed a lot for the coding efficient. The basic principle of predictive coding is predicting the current coding elements by those coded and constructed as they have some correlation, spatial of temporal. There exists two predictive coding in the standard: intra prediction and inter prediction based on single spatial and temporal corrrelation representatively. The thesis makes a deep research on combined inter-intra prediciton based on temporal-spatial correlation. The spatial corelation is higher for closer pixels and the neighboring regions of a smooth one may be smooth with a high probability. Based on the above hypothesis, the thesis proposes a novel combined inter-intra predictive tactics. It devides the current macroblock into inter-prediction region and intra-prediction one, the later is next to the reference macroblocks. And the experimental results show that the combined prediciton will get a better performance. The reduction of overhead bits of mode information and the selecting of weighting coefficients of inter-prediction and intra-prediction are the two technical breakthroughs.Although video coding decreases many bits needed to describe video signal, video frame may suffer some artifacts due to a block-based quantization during the compression. Blocking and ringing and two of the most visible artifacts, and the human is sensitive to the former artifact. Therefore, a high efficient algorithm to reduce blocking artifact is very necessary. There are a lot of compressed video de-blocking methods, for example, the in-loop de-blocking technology. K-SVD (K-Singular Value Decomposition) for image noise reduction is a method based on sparse representation over adaptive over-complete dictionary. The thesis integrates the K-SVD algorithm and video coding to reduce compressed video blocking artifact. The K-SVD consists of two stages, the sparse representation of the signal over a given dictionary and the update of the dictionary and the representation coefficients. The experimental results demonstrate the improvement of performance in both subjective and objective criteria. It gets about 0.4dB PSNR gain for I frame.