Distortion measure analysis for efficient block matching and face region based video coding
Date of Issue2009
School of Electrical and Electronic Engineering
Video coding has been intensively studied for over one decade and a few video compression standards have been well established with wide applications. These coding paradigms or standards have a number of commonalities, the most salient one among which is the hybrid coding structure consisting of diverse techniques to remove both the temporal and the spatial redundancy. One of the major techniques is block matching which has been used in inter-prediction like block motion estimation as well as in intraprediction such as intra-mode selection in H.264/AVC. Another important topic is bit allocation for rate control to optimize the coding performance. In this thesis, we investigate the block matching technique and face region priority based bit allocation for video coding. While the block matching based motion estimation has been widely studied, this thesis addresses the efficient block matching from the distortion measurement perspective. Specifically, three aspects on the distortion measurement analysis are considered, namely, a new multiplication-free distortion metric, transform-exempt sum of absolute Hadamard transformed differences (SATD) calculation, and a subblock-based distortion metric for efficient block matching, which are summarized in order as follows. It is known that the distortion measure function plays a pivotal role in both matching accuracy and computational complexity. To avoid multiplication operations for simpler implementation, sum of absolute difference (SAD) is normally taken as a substitute to approximate the most widely accepted benchmark, mean squared error (MSE). Although it is well known that SAD is an approximate to MSE, it is necessary to know how accurate the approximate is quantitatively, which can provide us more insights and guidance in selecting an appropriate matching criterion function and developing some new ones. However, there is a lack of such quantitative treatment to this fundamental question. In this thesis, we firstly examine the quantitative deviation of SAD from MSE. In order to reduce the deviation, a new matching criterion, namely weighted sum of absolute difference (WSAD), is thereafter proposed, which enhances the matching accuracy while still maintaining the desirable multiplication-free property. The proposed WSAD is also experimentally validated by applying it in block motion estimation for video coding, showing better rate-distortion performance than SAD. Secondly, in the latest video coding standard H.264/AVC, the sum of absolute Hadamard transformed differences (SATD) is a new distortion metric adopted as an alternative to the SAD to improve coding efficiency.
DRNTU::Engineering::Electrical and electronic engineering::Electronic systems::Signal processing