Main Research Activities

Fused Data Visualization	Large Scale Data Visualization	Spiral Architecture
Visual Affective Computing	Visual Surveillance	Visualizing Citation & Co-authorship Networks

Fused Data Visualization

Due to the rapid growth of information analysis and data mining technologies, the massive data sets available for access have been filled into multi-level information, including raw data and all kinds of knowledge (analytic results). As data sources become more and more complex, the current visual analytical tools are no longer satisfying the needs for exploring and analyzing the data.

This is mainly because that these tools are too simple to deal with the current complex multi-level data spaces. This situation raises the challenges in the current state of information visualization:

Due to the complexity of information, a single visual metaphor sometimes cannot well model the multi-level data that to be viewed in a single visualization.
Each existing visualization method (or algorithm) has its own limitations, in terms of satisfying domain specific requirements, when deal with complex data sets. These limitations reduce seriously the adaptability and quality of the method.

The proposed fused visualization methodology attempts to address the above problems by combining two or more existing visualization methods into a single visualization. It takes the advantages and reduces the weaknesses of these existing methods. We have successfully applied this methodology into each stage of our proposed Analytic Information Visualization (AIV) pipeline.

We seek ways to support the understanding of analyzed information through the use of appropriate visual metaphors, and the development of efficient layout and navigation methods.

Large Scale Data Visualization

The general aim of this research is to investigate interactive visualization solutions to the problem producing and displaying a complete global view of large information spaces in a limited screen resolution, and the problem of navigating through such large global views to find particular data items. Such an interactive visual representation of information will enable users not only to see the overall structure of a large information space with thousands (or tens of thousands) of data items on an ordinary computer screen, but also to easily and quickly navigate through the visualization to find particular information they want. The objectives of this research are:

To develop the theory of space-efficient graph visualization to provide the foundation of layout algorithms to be developed in this research.
- To investigate graph layout algorithms which enable the display of many items on a screen by maximising the utilization of two-dimensional space.
- To optimized the graph layout algorithms to significantly reduce the computational complexity of calculating the geometrical layout of large graphs with thousands (or tens of thousands) of nodes
To investigate an optimized navigation mechanism which will form the foundation of the navigation methods and algorithms to be developed.
- To investigate alternative focus+context navigation methods that can produce a sequence of large and high quality navigational view for data manipulation
- To investigate efficient algorithms that can significantly reduce the computational cost in view transformations during the navigation of large scale visualization.

Recent Grants

UTS internal research grant (Research Excellence Grant scheme), 2004, $20,000
ARC Discovery Project DP0665463, 2006-2008, $160,000

Spiral Architecture

This research activity aims to design and implement a robust surveillance system within the Spiral Architecture, a relatively new image and computationally powerful data structure for image representation. This system can fast detect moving objects and quickly return an accurate recognition of the observed objects in real-time. The objectives of this project are:

To correctly locate moving objects. This is to tell where the location of a moving object is in an image. An example is to locate a car moving in a given scene.
To fast and accurately detect moving objects. An example is to classify the model of the moving car in the scene.
To fast and accurately identify moving objects. An example is to identify the moving car by recognition of the car number plate.
To set up a small computer network of personal computers (PCs), called a cluster of PCs, for simultaneous (or parallel) and high-speed image processing including image segmentation, detection of image edge pixels, extraction of object contours, and recognition of objects.
To develop an interface between the cluster and a single digital video camera that will be used to capture moving objects.

An ARC Discovery Grant application will be submitted in 2003 by Dr. Xiangjian He. The estimated length of the project is three years. The associated researchers involved in this activity will mainly be Xiangjian He, Tom Hintz, and Qiang Wu; the other associated researchers might also be involved at a later stage.

Visual Affective Computing

This research activity aims to enable intelligent interaction between humans and computers through the development of vision-based systems that own human-like perceptual computing abilities. In accordance with this aim, we have been working on two novel areas within the field: facial beauty analysis and multi-modal emotion recognition.

Facial Beauty Analysis - This research has tackled the problem of facial beauty analysis by creating a database of female faces and conducting a survey on how human observers perceive and scale facial beauty. The next step in this research was (a) building an intelligent system that can perform automatic beauty classification by extracting the facial features from the static images; (b) calculating various proportions based on the Golden Ratio and Facial Thirds; (c) training a classifier to mimic the human decision making on facial beauty and evaluating its performance.

Multi-modal Emotion Recognition - This research aims to build a bimodal/multimodal system that can extract features from expressive face and upper-body gestures using computer vision and image processing techniques and exploit such features in machine learning and pattern recognition approaches for affect recognition. We achieve bimodal emotion recognition from facial and upper-bodily expression by exploring data acquired in laboratory settings. We further fuse the face and upper-body modalities by exploring and comparing appropriate fusion approaches. As an integral part of this research, we created a novel bimodal face and body gesture database (FABO) suitable for use in automatic vision-based analysis of human nonverbal communicative behavior. The FABO database contains videos of face and body expressions recorded by the face and body cameras, simultaneously. This database is the first to date to combine facial and body displays in a truly bimodal manner, hence enabling significant future progresses in affective computing research.

Visual Surveillance

Computer vision-based video surveillance aims to provide computer assistance to the human analysis of surveillance footage that could be used in real systems. It involves many different techniques of computer vision to analyse the shape motion and other appearance features of the surveillance footage. Object tracking is seen as one of the more useful aspects of this research with people tracking being a more complicated subset of the general tracking problem as humans are deformable objects that are also prone to self-occlusion. Human environments also tend to be less structured and more crowded than other environments increasing the difficulty of object and feature extraction. Surveillance systems of such environments are also likely to consist of many camera views that are disjointed, often significantly, making object motion cues unavailable for parts of the path. Such scenarios are common because of the cost of acquiring enough equipment to provide full coverage, especially with high enough resolution to measure accurate biometric information.

A video surveillance research project within the Computer Vision Research Group (CVRG) at UTS looks at building on this existing surveillance scenario to track individuals by matching appearance features between cameras. This is done using session-based biometrics where a surveillance session is defined as the segment of time from when an individual enters a building`s surveillance system, moves around inside the surveillance area and then exits. A typical surveillance session would be a portion of one day as people enter the workplace to conduct their business before leaving. Thus features such as clothing colour and clothed height would tend to remain constant throughout a single surveillance session, even if they might not be consistent over a longer period as people often change their clothes and shoes from day to day. Such features are not enough to necessarily identify a person as might be expected from fingerprints; however it can be used to distinguish between people. By analysing enough features the matches between individuals as they move about can obtained to a high degree of confidence. The current research is based upon fusing the information of height and colour features to provide a base framework for the use of track matching in generating disjoint camera matching.

Recent Grants CVRG has recently been awarded the following competitive research grants to support and extend their video surveillance activities:

UTS internal research grant (Research Excellence Grant scheme), 2003, $18,000
ARC Discovery Project DP0452657, 2004-2006, $168,000
AusIndustry Start Graduate, 2004-2005, $16,274
ARC Linkage Project LP0668325, 2006-2008, $354,000

Visualizing Citation & Co-authorship Networks

The aim of this research is to develop a collective visualization system that collates and interprets all relevant citation data and then presents it in simple, interactive, ease-to-use and understandable graphical formats. The main objectives are:

Investigate effective techniques for visualizing citation/co-citation network.
Investigate optimized graphic methods and layout algorithms.
Investigate dynamic visualization of co-citation networks of a specific domain in corresponding to the updates of bibliographic data.
Develop a visualization prototype for evaluating individual��s academic performance that provides scientists with a roadmap of their own research field.
Define the methodologies and tools for future researches.