Research Students/Assistants/Visiting Scientists
back to my
home page
Student/Research Assistant Vassilis Emmanuel Seferidis , PhD (1990-1993)
Title: Video
Coding for ATM Networks
Funding: British Telecom
Research Assistant: Jianping
Xiong (1990-1993)
Title: Interactions of real-time speech and video in an ATM environment
Funding: EPSRC
Student: Jian Azari, PhD (1990-1993)
Title: Transmission of Mixed Video,
Speech and Data Services over ATM Networks
Student: Tan Kuan Hui, MSD (by Dissertation) (1992-93)
Title: Low Bit
Rate Video Coding using An Adaptive 1D/2D DCT Technique
Student: Demetrios G. Sampson, PhD (1990-1995)
Title: Lattice
Vector Quantization for Image Coding
Student: Eduardo Antônio Barros da Silva, PhD (1991-1995)
Title: Wavelet
Transforms for Image Coding
Student: Anastasios Dagiuklas, PhD (1992-1995)
Title: Modeling and Traffic Management of Video Services in an ATM Network
Student: Sérgio Manuel Maciel de Faria, PhD (1993-1996)
Title: Very Low Bit Rate Video Coding using Geometric Transform Motion Compensation
Research Assistant: Qi Wang (1993-1996)
Title: Coding of Super High definition images
Funding: EPSRC
Student/Research Assistant: Richard Aldridge, PhD (1994-1996)
Title: Continuous Quality Assessment of Digitally-Coded Television Pictures
Funding: Independent Television Commission/EU MOSAIC project
Student/Research Assistant: David Hands, PhD (1994-1997)
Title: Mental Processes in the Evaluation of Digitally-Coded Television Pictures
Funding: Independent Television Commission/EU MOSAIC project
Student: Pedro António Amado de Assunção, PhD (1994-1997)
Title: Video Transcoding
for Matching Network Bandwidth Constraints
Student: David Wilson, PhD (1994-1997)
Title: Efficient Coding and
Transmission of Two-Layer Video
Student: Joaquín García-Hernández, PhD (1994-1999)
Title: Management of the Quality of
Video Services in ATM Networks
Student: Ali Dawood, PhD (1995-1999)
Title: Content based processing and
modeling of MPEG-1 video streams
Student: Fernando Lopes, PhD (1996-2000)
Title: Motion Estimation for Very Low Bitrate Video Coding
Student/Research Assistant: Kwee Teck TAN, PhD (1996-2000)
Title: Objected Picture Quality Measurement for MPEG-2 coded Video
Funding: Independent Television Commission/ EU, TAPESTERIES project
Student: Damianos Gavalas, PhD (1997-2001)
Title:
Funding:
Student/Research Assistant: Eva Rosdiana, MSD (By dissertation) (1998-2000)
Title: Transmission of Transcoded video over ABR Networks
Funding: British Telecom
Student: Marcos Paredes Farrera, PhD (2001-Now)
Title: Measurement and Analysis of Internet Traffic: Based on a Packet-by-Packet Methodology and Flow Classification
Title: An Efficient Bandwidth Reservation Scheme For VBR Video Streaming Over The Internet
Research Assistant: Ernesto
L. Andrade Neto (2002-2004)
Title: Multi-platform
e-publishing for leisure and Interactive sports advertising
Funding: EU, MELLISA
project
Student: Irwan Prasetya Gunawan, PhD (2002-Now)
Title: Video Quality
Assessment in the absence of full-reference information
Funding: BTExact
Student/Research Assistant: Mahdi Ghandi, PhD (Jan 2002-NOW)
Title: video coding for wireless communication
Funding: EPSRC
Student: Huihai Lu, PhD (2002-Now)
Title: Object recognition and tracking
Data, Speech and Video are the three main categories of service that will dominate future telecommunications. Video being the newest service, a description of video coding techniques is presented with the introduction of two-layered coding which offers network operators flexibility during overload. Statistics of variants of this coding technique are investigated, and a two-layered coder called CBR-VBR-b is selected for its superior performance.
Asynchronous Transfer Modes are described which define the characteristics
of networks that have to cope with new services. The Orwell protocol is
selected for modification, to control medium access and scheduling of services
to a ring Local/Medium Area Network, maintain Quality of Service at all
times. These modifications included the concept of bandwidth donation,
and a new bandwidth allocation algorithm. The final protocol, named
Orwell-D, is finally compared to a counterpart call Magnet-II and common traits
are revealed as a prerequisite for these types of networks.
go to top
An adaptive one- or two-dimensional Discrete Cosine Transform (1D/2D DCT) technique is conceptualised and demonstrated to be a more appropriate strategy for coding motion compensated prediction error images. The indiscreet application of 2D-DCT on pulsive linelike components will scatter the mage's energy in the transform domain. This not only hinders efficient entropy coding of the coefficients but also lowers the resilience of the transform coder to quantization noise.
There are two key components to the proposed adaptive DCT codec. A
block classifier and a correspondingly matched entropy coder.
Two classification parameters were introduced for the block classifier.
One was structured after the first-order correlation coefficient weighted by
the variance while the other is variance-based. A systematic evaluation
procedure of the classifiers' performances was also formalised.
A significant finding was that in a real MCP error image, the number of blocks
which are classifiable as 1D-blocks is predictable.
The adaptive coder was proposed with two bit-allocation schemes. The
final version settles with a fixed-overhead to differentiate the type of
transform used. The 1D/2D DCT coder with a variance-based classifier is
recommended. Improvements against the homogenous 2D-DCT system both in
terms of peak signal to noise ratio and subjective assessments were
achieved. Observable ringing effects along edges, which is usual in
conventional transform coding, was reduced.
go to top
This thesis investigates the application of lattice vector quantization to video data compression. Lattice Vector Quantization (LVQ) is a special category of vector quantization (VQ), where the VQ codebook is designed using a finite set of lattice points. The main motivation for investigating lattice VQ is that fast algorithms for the selection of the best-matched code vector can be developed by exploiting the geometric properties of regular lattices.
A product-code lattice vector quantization scheme is first investigated. Two categories of lattice codebooks, namely, single-shell and multi-shell codebooks, are built using the innermost shells of 8- and 16-dimensional lattices. Algorithms are sub-optimal, in the sense that they do not ensure that the best available code vector is always selected. However, they are designed to offer significantly reduced complexity as compared to the optimum exhaustive full codebook search. Experiments are carried out to evaluate the computational load and the coding accuracy of these algorithms. Finally, the performance of various lattice codebooks is tested and compared with the orientation codebooks designed using the Generalized Lloyd's Algorithm. Simulation results for image sequence coding illustrate that lattice codebooks achieve similar PSNR performance with GLA-designed codebooks, provided that an efficient entropy coder is incorporated.
A multi-stage lattice vector quantization method referred to as Successive
Approximation Lattice Vector Quantization (SA-LVQ) is then studied. The
basic idea in SA-LVQ is that the original input vector is successively refined
using vectors of progressively decreasing magnitude and a finite set of
prototype orientations. Criteria for the selection of the orientation
codebook are devised and the parameters related to the SA-LVQ task are computed
for lattice codebooks based on the inner-most shells of D4, E8 and D16.
SA-LVQ is incorporated with (i) wavelet transform
(ii) zero-tree prediction, and (iii) arithmetic coding, for the development of
an image coding scheme, referred to as Successive Approximation Wavelet Lattice
Vector Quantization (SA-W-LVQ). The application of SA-W-LVQ to low bit
rate video coding is then examined. The video codec presented consists of
two main elements: (i) the motion
estimation/compensation which is performed using Overlapped Block Matching
(OBM), and (ii) the coding of the motion compensated prediction error using
SA-W-LVQ. The overall performance of the OBM-SAWLVQ video codec is
examined for low bit rate coding and compared with other wavelet/subband video coding methods and the RM.8 model of the ITU
standard H.261 video codec.
go to top
SNR Scalability (SNRS) and Data Partitioning (DP) have both been proposed by MPEG-2 as a means of guaranteeing minimum image quality in the presence of information loss during ATM transmission of video services. The advantage of SNRS is its ability to provide superior base layer image quality to that of DP while its main drawback is a significant increase in total bit-rate compared to its single-layer counterpart.
The thesis identifies the enhancement layer as the source of this inefficiency, from which two approaches to its improvement are considered. In the first, an arithmetic coder scheme is proposed as an alternative approach to enhancement layer coding. Several arithmetic coding schemes are compared experimentally with the best of them delivering up to 20-25% better enhancement layer compression. In the second, an optimisation strategy is proposed in which DCT coefficient levels are 'adjusted' in order to minimise bit-rate for a given target distortion. With up to 15% reductions in the total bit-rate, the optimised SNRS coder is thus able to demonstrate comparable coding efficiency to DP as well as single-layer coders.
ATM transmission aspects pertaining to prioritisation and bandwidth efficiency are also investigated for VBR MPEG. The shortcomings of existing priority schemes, are pointed out and an alternative two-layer frame partitioning scheme proposed in order to provide better protection to anchor frames and improved statistical multiplexer gain, without incurring any bit-rate overhead.
Finally, the introduction of an appropriate delay between base and
enhancement layers, is shown to be effective in decorrelating
SNRS layers prior to VBR transmission. For different multiplexer
simulation scenarios, decorrelated SNRS sources are
able, despite bit-rate overheads, to outperform or at least match single-layer
transmission cell loss performance over a broad range of coding influences
including varying frame alignments, varying base layer bandwidths and the
application of traffic shaping.
go to top