Please use this identifier to cite or link to this item:
Title: Enhancing techniques for a standard conforming real-time video codec
Authors: Lai, Kam-cheong
Degree: M.Phil.
Issue Date: 2002
Abstract: The ultimate goal of video coding is to achieve a balance between the spatial quality and the temporal quality subject to human judgment with the best coding efficiency and the least computational requirement. Since inter-operability is of paramount importance, several video coding standards have been established such as H.261, H.263, MPEG-1, MPEG-2, and MPEG-4. Among these standards, the most widespread coding structure takes the form of combing a motion-compensation with Discrete Cosine Transform (DCT). Since standards only specify the decoder's syntax, it provides flexibility for improvement. In this thesis, we present new standard- conforming video coding techniques, which can offer better coding efficiency. Efficient motion estimation is essential for achieving good quality video coding. With studying the distribution of the motion vectors obtained by the Full Search algorithm, we optimize the prediction process of the motion estimation by classifying the blocks into four different classes based on a newly proposed group of reference motion vector. Different search patterns and strategies are used for different classes. The proposed algorithm improves the computational requirement and coding efficiency when compared to other fast motion estimation algorithms. Besides, a Weighted Sum of Absolute Difference (WSAD) matching criterion function for motion estimation is proposed. WSAD is designed based on the characteristic of the DCT. WSAD enhances the coding efficiency while the computational overhead is very small. To achieve better video quality in very low bit rate situation, an object-based video coding scheme is proposed. Each frame of a video is segmented into three non-overlapping regions, namely background, non-facial foreground and facial region. Smaller quantization step (better quality) is used for the facial region while a larger quantization step is used for the background to improve the coding efficiency. However, the non-facial foreground is kept in normal coding quality to prevent degradation of important regions other than facial region such as hand movement. The proposed method is real-time ready. Since the video data encoded in block-based video codec has an inherently variable bit rate, a rate control algorithm is proposed in this thesis. As human judges the quality of the output, we adopt the information of human visual system to our rate control system. The proposed algorithm is able to adjust the operating point of the encoder automatically to a balance region between the spatial and temporal quality. Besides, experimental results show that the proposed rate control algorithm prevents motion jerkiness in condition of bandwidth fluctuation. Finally, we proposed a fast direct spatial filtering algorithm on DCT-based compressed video. As video signals are often compressed for transmission or storage, manipulation of compressed video signals is required. Filtering is a technique often used in these processes. Direct filtering on DCT domain can speedup the filtering process without going through the inverse transform of data from DCT domain to spatial domain. Our proposed method although performs spatial filter and inverse DCT transform at the same time, the computational requirement is less than a fast inverse DCT transform.
Subjects: Hong Kong Polytechnic University -- Dissertations
Video compression
Digital video
MPEG (Video coding standard)
Pages: xiii, 110 leaves : ill. ; 30 cm
Appears in Collections:Thesis

Show full item record

Page views

Last Week
Last month
Citations as of Jun 4, 2023

Google ScholarTM


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.