Optimized techniques for video compression and enhancement

Abstract

Today, with the rapid development of cost-effective and popular digital recording devices, digital video contents are being generated and distributed at a phenomenal rate. As the volume of digital video is notoriously huge, efficient techniques to process this large amount of data are also becoming increasingly important. The requirements for efficient video transmission and storage make video compression very important in video processing to reduce the data rate. Meanwhile, to support universal multimedia access over heterogeneous networks and devices, video transcoding that processes an existing compressed video to better suit new application constraints has also attracted much attention. To meet the growing demand for better and faster visual communication services, optimized techniques for these processes have been explored in order to achieve better performance in terms of complexity, quality, and bit rate. This requirement in turn becomes challenging due to the trade-off among these measures. Owing to the scarce resources and constraints encountered in real-time applications, video encoder and transcoder may need to reduce the complexity at the expense of degraded visual quality. When the resources and constraints are no longer concerned, how to obtain better quality from these encoded videos becomes important. To address this problem, quality enhancement by post-processing has been employed and become an active research topic in recent years. In this thesis, we propose new and efficient techniques to address key problems in video ATTENTION: The Singapore Copyright Act applies to the use of this document. Nanyang Technological University Library Contents vii compression and enhancement. We first focus on optimizing the video coding process, which aims to reduce the computational complexity while maintaining an acceptable visual quality, in order to make it suitable for real-time applications. Motion estimation is mostly considered in our solution due to it being the most computationally intensive part of a typical video encoder. New block-matching measures are proposed to reduce the computational complexity of block-based motion estimation. These proposed measures evaluate the match between two blocks by making use of the features such as block sum and block variance, which can be easily computed from integral frame attributes with optimal number of computations. Next, a complete picture of efficient format transcoding and downsizing transcoding between the existing compression standard H.263 and the stateof-the-art H.264/AVC standard is presented. To speed up the process, fast motion vector re-estimation and intra-prediction mode selection are proposed. Furthermore, an enhanced rate control is employed for H.264/AVC transcoding to improve the transcoded video quality. In particular, a model is developed to approximate the relationship between the total number of coding bits and quantization step sizes of the precoded and transcoded videos, which is used for selecting the quantization parameters at the sequ...