Computer Graphics/Multimedia/Image Processing

1. Fuzzy Data Fusion for applications in Computer Graphics and Virtual Reality
2. Centred Form-based Shape Representation and Solid Modeling
3. Near-lossless Image Compression
4. Intelligent Knowledge-based Real-time Imaging: Robust and Parallel Image Matching on Distributed Systems

Fuzzy Data Fusion for applications in Computer Graphics and Virtual Reality
(Investigator: Keith C.C. CHAN)
Data fusion is concerned with the combining of data collected from multiple sources to support intelligence generation. Many real-world problems can be better solved if data from similar and dissimilar sources can be combined in some meaningful ways. For example, robotic vehicle control systems can better emulate biological situation development process by combing outputs from multiple on-board sensors: earth resource studies can better correlate satellite-based multispectral imagery with map-based information; airport safety system may be able to fuse ground-based Doppler radar data with range-finding sensors to identify potentially dangerous flight profiles, etc. For this project, our focus is on the development of a fuzzy login based image fusion technique in the superimposing of computer-generated images over an observer's view of the real world so that properly registered 3D computer graphics can be displayed with the environment the observer is in. Such a technique will be used in applications such as telemedicine, remote-guided surgical navigation systems or virtual medical systems. By fusing live endoscopic video data with synchronously recorded 3D sensor data in real time, the spatial position of instruments relative to anatomical structures and the access paths to the regions of interest can be visualized.

Centred Form-based Shape Representation and Solid Modeling
(Investigator: Paul G.BAO)
Shape representation is one of the fundamental problems in computer graphics. The applications of shape representation can be found in various fields such as Multimedia, Computer Animation, Virtual Reality, Computer Vision, etc. the forms of shape representation such as deformations, implicit and parametric shapes, CSG have been widely studied and algorithmic approaches to the shape representation such as piecewise parametric polynomials, algebraic surfaces, implicit and parametric shapes have been available and widely used. However, these representation techniques normally involve substantial computation and introduce approximation errors. The problems of shape representation can be described using two generic algorithms based on interval analysis and inclusion functions techniques. Using the algorithmic centered forms techniques, an interval analysis-based systematic shape representation technique will be produced. This technique not only solves the problems in shape representation algorithmically with high efficiency but also controls the error approximations.

Near-lossless Image Compression
(Investigator: Paul G, BAO)
In many important applications, such as medical imaging and remote sensing, users desire lossless image compression because of their stringent demands on the faithfulness of the reconstructed image to the original source. Unfortunately, the achievable lossless compression ratios are quite low, typically being 2 to 3. Furthermore, there are both theoretical and practical indications that recent sophisticated lossless image coders have reached the level of compression that is very close to the theoretical limit. Therefore to achieve better compression ratios while satisfying the stringent quality requirements of many sensitive applications, the new JPED standard being developed has the provision for so-called near-lossless compression with a user-specified bound on the error magnitude. In this proposal, aiming at enhancing the performance of the near-lossless compression scheme based on the new multilevel weighted finite automata transformation (MWFA) and probabilistic multilevel weighted finite automata transformations (PMWFA), hybridized with Context-based. Adaptive techniques, wavelet, adaptive residue quantization, nonlinear gradient-adjusted predictor and error modeling, and context formation and quantization.

Intelligent Knowledge-based Real-time Imaging: Robust and Parallel Image Matching on Distributed Systems
(Investigators: James N.K. LIU, Jane YOU (Griffith University, Australia), Edwige Pissalous (Universite Paris XI, France))
Automatic recognition of any object involves several stages such as low level processing, model formation, features extraction and model matching. Those algorithms and procedures that are now available for the above stages tend to be highly complex, context dependent and very difficult for use by non-experts. This project aims at tackling the above limitations to develop a semantically integrated system in support of automating the recognition of objects at various stages. It will investigate and develop a real-time image matching mechanism via parallelism on distributed memory multicomputers at low cost. By introducing a dynamic image feature extraction procedure and a hierarchical image matching scheme, the intelligent system is expected to provide means of recognizing and localizing the specific object in textured images invariant of translation, rotation and scale changes with respect to robustness, reliability and efficiecy. The developed technology has applications to a wide range of areas including monitoring soil and land use, mining and mineral processing, handling of agricultural products, inspecting manufactured products, medical diagnostics, and analyzing biological or aerial imagery.