11 |
Transform Coefficient Thresholding and Lagrangian Optimization for H.264 Video Coding / Transformkoefficient-tröskling och Lagrangeoptimering för H.264 VideokodningCarlsson, Pontus January 2004 (has links)
H.264, also known as MPEG-4 Part 10: Advanced Video Coding, is the latest MPEG standard for video coding. It provides approximately 50% bit rate savings for equivalent perceptual quality compared to any previous standard. In the same fashion as previous MPEG standards, only the bitstream syntax and the decoder are specified. Hence, coding performance is not only determined by the standard itself but also by the implementation of the encoder. In this report we propose two methods for improving the coding performance while remaining fully compliant to the standard. After transformation and quantization, the transform coefficients are usually entropy coded and embedded in the bitstream. However, some of them might be beneficial to discard if the number of saved bits are sufficiently large. This is usually referred to as coefficient thresholding and is investigated in the scope of H.264 in this report. Lagrangian optimization for video compression has proven to yield substantial improvements in perceived quality and the H.264 Reference Software has been designed around this concept. When performing Lagrangian optimization, lambda is a crucial parameter that determines the tradeoff between rate and distortion. We propose a new method to select lambda and the quantization parameter for non-reference frames in H.264. The two methods are shown to achieve significant improvements. When combined, they reduce the bitrate around 12%, while preserving the video quality in terms of average PSNR. To aid development of H.264, a software tool has been created to visualize the coding process and present statistics. This tool is capable of displaying information such as bit distribution, motion vectors, predicted pictures and motion compensated block sizes.
|
12 |
On Non-Convex Splitting Methods For Markovian Information Theoretic Representation LearningTeng Hui Huang (12463926) 27 April 2022 (has links)
<p>In this work, we study a class of Markovian information theoretic optimization problems motivated by the recent interests in incorporating mutual information as performance metrics which gives evident success in representation learning, feature extraction and clustering problems. In particular, we focus on the information bottleneck (IB) and privacy funnel (PF) methods and their recent multi-view, multi-source generalizations that gain attention because the performance significantly improved with multi-view, multi-source data. Nonetheless, the generalized problems challenge existing IB and PF solves in terms of the complexity and their abilities to tackle large-scale data. </p>
<p>To address this, we study both the IB and PF under a unified framework and propose solving it through splitting methods, including renowned algorithms such as alternating directional method of multiplier (ADMM), Peaceman-Rachford splitting (PRS) and Douglas-Rachford splitting (DRS) as special cases. Our convergence analysis and the locally linear rate of convergence results give rise to new splitting method based IB and PF solvers that can be easily generalized to multi-view IB, multi-source PF. We implement the proposed methods with gradient descent and empirically evaluate the new solvers in both synthetic and real-world datasets. Our numerical results demonstrate improved performance over the state-of-the-art approach with significant reduction in complexity. Furthermore, we consider the practical scenario where there is distribution mismatch between training and testing data generating processes under a known bounded divergence constraint. In analyzing the generalization error, we develop new techniques inspired by the input-output mutual information approach and tighten the existing generalization error bounds.</p>
|
13 |
Scalable video compression with optimized visual performance and random accessibilityLeung, Raymond, Electrical Engineering & Telecommunications, Faculty of Engineering, UNSW January 2006 (has links)
This thesis is concerned with maximizing the coding efficiency, random accessibility and visual performance of scalable compressed video. The unifying theme behind this work is the use of finely embedded localized coding structures, which govern the extent to which these goals may be jointly achieved. The first part focuses on scalable volumetric image compression. We investigate 3D transform and coding techniques which exploit inter-slice statistical redundancies without compromising slice accessibility. Our study shows that the motion-compensated temporal discrete wavelet transform (MC-TDWT) practically achieves an upper bound to the compression efficiency of slice transforms. From a video coding perspective, we find that most of the coding gain is attributed to offsetting the learning penalty in adaptive arithmetic coding through 3D code-block extension, rather than inter-frame context modelling. The second aspect of this thesis examines random accessibility. Accessibility refers to the ease with which a region of interest is accessed (subband samples needed for reconstruction are retrieved) from a compressed video bitstream, subject to spatiotemporal code-block constraints. We investigate the fundamental implications of motion compensation for random access efficiency and the compression performance of scalable interactive video. We demonstrate that inclusion of motion compensation operators within the lifting steps of a temporal subband transform incurs a random access penalty which depends on the characteristics of the motion field. The final aspect of this thesis aims to minimize the perceptual impact of visible distortion in scalable reconstructed video. We present a visual optimization strategy based on distortion scaling which raises the distortion-length slope of perceptually significant samples. This alters the codestream embedding order during post-compression rate-distortion optimization, thus allowing visually sensitive sites to be encoded with higher fidelity at a given bit-rate. For visual sensitivity analysis, we propose a contrast perception model that incorporates an adaptive masking slope. This versatile feature provides a context which models perceptual significance. It enables scene structures that otherwise suffer significant degradation to be preserved at lower bit-rates. The novelty in our approach derives from a set of "perceptual mappings" which account for quantization noise shaping effects induced by motion-compensated temporal synthesis. The proposed technique reduces wavelet compression artefacts and improves the perceptual quality of video.
|
Page generated in 0.19 seconds