Weighted Combination of Sample Based and Block Based Intra Prediction in Video Coding

Abstract: The latest standard within video compression, HEVC/H.265, was released during 2013 and provides a significant improvement from its predecessor AVC/H.264. However, with a constantly increasing demand for high denition video and streaming of large video files, there are still improvements that can be done. Difficult content in video sequences, for example smoke, leaves and water that moves irregularly, is being hard to predict and can be troublesome at the prediction stage in the video compression. In this thesis, carried out at Ericsson in Stockholm, the combination of sample based intra prediction (SBIP) and block based intra prediction (BBIP) is tested to see if it could improve the prediction of video sequences containing difficult content, here focusing on water. The combined methods are compared to HEVC intra prediction. All implementations have been done in Matlab. The results show that a combination reduces the Mean Squared Error (MSE) as well as could improve the Visual Information Fidelity (VIF) and the mean Structural Similarity (MSSIM). Moreover the visual quality was improved by more details and less blocking artefacts.