High Efficiency Prediction Methods for Current and Next Generation Video Coding

Abstract

Consumption and production of video signals drastically changed in recent years. Due to the advances in digital consumer technology and the growing availability of fast and reliable internet connections, an increasing amount of digital video sequences are being produced, stored and shared every day in different parts of the world. Video signals are inherently larger in size than other types of multimedia signals. For this reason in order to allow transmission and storage of such data, more efficient compression technology is needed. In this thesis novel methods for enhancing the efficiency of current and next generation video codecs are investigated. Several aspects of interest to video coding technology are taken into account, from computational complexity and compliance to standardisation efforts, to compression efficiency and quality of the decoded signals. Compression can be achieved exploiting redundancies by computing a prediction of a part of the signal using previously encoded portions of the signal. Novel prediction methods are proposed in this thesis based on analytical or statistical models with the aim of providing a solid theoretical basis to support the algorithmic implementation. It is shown in the thesis that appropriately defined synthetic content can be introduced in the signal to compensate for the lack of certain characteristics in the original content. Some of the methods proposed in this thesis aim to target a broader set of use cases than those typically addressed by conventional video coding methods, such as ultra high definition content or coding under high quality conditions.