ICTACS 2006

PREFACE.

CONTENTS.

Graph algorithms have several important applications in fields such as computer science, software engineering, mathematics, engineering, business, and bioinformatics. Researchers and practitioners in these disciplines often need to experiment with empirical data about graphs to gain deeper insights into their properties, which may lead to general proofs. Students also need to learn and be able to implement graph algorithms for their applications. However, these individuals often have varying backgrounds and training, and they may not have a working knowledge of programming tools to implement graph algorithms. The goal of our research is to create a software system which allows users to easily create and implement graph algorithms through a simple graphical user interface, without any coding. Towards this goal, we have developed several systems that can be used to draw and manipulate graphs as well as to execute graph algorithms. The development of the general system raises a number of interesting research questions that we will discuss. To illustrate the need for experimentation with different graph algorithms, we describe some examples of our research in different areas of graph theory and computational geometry. We also describe some features of our systems that we have found useful in teaching graph theory.

The keyword list-based spam email detection system uses keywords in a blacklist to detect spam emails. To avoid detection, keywords are written as misspellings, for example "virrus", "vi-rus" and "viruus" instead of "virus". The system needs to update the blacklist from time to time to detect spam emails containing such misspellings. However it is impossible to predict all possible misspellings for a given keyword to add those to the blacklist. This paper proposes a statistical framework to solve this problem. A keyword is represented as a Markov chain where letters are states. A Markov model is then built for the keyword. In order to decide an unknown word as a misspelling of a given keyword, a statistical hypothesis test is used. Experiments showed that the proposed statistical models could achieve the detection error rate of 0.1%.

This paper deals with learning cooperation in a group of predators chasing a prey. A learning predator has no knowledge about the environmental information. It only knows that it should adapt its actions to other predators' expectations. The learning classifier system is a very interesting tool for the machine learning technique. Indeed, it helps a learning agent to evaluate its actions and to take right actions corresponding to situations in order to have a better cooperation with other agents. The results show how the learning is constructed in such collective task.

Retrieving relevant parts of a meeting or a conversation recording can help the automatic summarization or indexing of the document. In this paper, we deal with an original task, almost never presented in the literature, which consists in automatically extracting questions utterances from a recording. In a first step, we have tried to develop and evaluate a question extraction system which uses only acoustic parameters and does not need any ASR output. The parameters used are extracted from the intonation curve and the classifier is a decision tree. Our first experiments on French meeting recordings lead to approximately 75% classification rate. An experiment in order to find the best set of acoustic parameters for this task is also presented in this paper. Finally, data analysis and experiments on another French dialog database show the need of using other cues like the lexical information from an ASR output, in order to improve question detection performance on spontaneous speech.

We propose a simple but reliable method to filter a collection of images of some object category that might contain many outlier images. Such collections are straightforward to obtain, for example, by using Google Image Search. To retrieve most of objects relevant to the category, visual information consistency is used. We use SIFT (Scale Invariant Feature Transform), a state of the art method for representing images with descriptors which are invariant to scale and orientation, to compute the similarity between two images and form a graph of which nodes are images and edges connected by two nodes are weighted by the similarity between two associated images to describe the consistency degree. Then, a subgraph is extracted by a greedy densest component algorithm to return the result images. The proposed approach has been tested on different categories and shown promising results.

This paper presents a method for modeling objects at multiple scales using a graph-based model and a strategy for matching of two structure hierarchies. The visual features such as ridge and peak detected at certain scales correspond to a node in an Attributed Relational Graph (ARG). Edges are inserted between nodes on two consecutive scales based on how the region associated to the feature at one scale is covered by the region associated to the other feature at the next scale. Using this representation, the process of object recognition is expressed as a matching problem of two graphs, known as the problem of searching for the maximal sub-graph isomorphism in graph theory. However graph matching is a highly complex and time-consuming. For matching two attributed relational graphs, the dissimilarity measure between two nodes is computed. In order to reduce the search space of the algorithm, symbolic constraints based on feature types are proposed. This technique is simple and well-adapted to our object model. We demonstrate our approach on an image database with some variations of object in images such as rotation, translation and small resolution change.

Nowadays, we are living in the content-based image retrieval (CBIR) age. The users would like to give the semantic queries, but the semantic understanding of images remains an important research challenge for the image and video retrieval community. We have approached the CBIR at semantics level by using visual information concepts (VIC) and automatic image annotation (AIA). We have linked the semantic concepts to the image at two levels, the common level and the private level. In the common level, we used the VIC and linking automatically VIC to image data based on the priori knowledge. In the private level, we performed the AIA based on the cross-media relevance model with some improvements. The content image retrieval process is based on the comparison of the intermediate descriptor values in VIC associated with both the semantic data and the image data. Irrelevant images are rejected and the remaining images are ranked by AIA. Our experiment results have shown that the performance of our system is better in the meanings of precision and recall than the traditional systems only based query images or only based on VIC or AIA.

In this paper, we are presenting a model for multimodal content analysis. We are distinguishing between media and modality, which helps us to define and to characterize 3 inter-modal relations. Then we are applying this model for recorded course analysis for e-learning. Different useful relations between modalities are explained and detailed for this application. We are also describing on two other applications: telemonitoring and minute meetings. Then we compare the use of multimodality in these applications with existing inter-modal relations.

This paper presents the method of multi-resolution analysis used in 2D image data to extract the curved edge features. The method is based on the combination of multi-resolution decomposition through Wavelet Packet and Prime Ridgelet transform. We call this combination Prime Wavelet Packet Contourlet Transform-PWPC. At each leave of Packet Wavelet Packet Tree, the prime ridgelet transform is applied on the band pass image or packet, which contains the high frequency data. The experiment shows that the PWPC coefficients are good approximations to curved edges. The speed of PWPC is faster than that of the basic Curvelet transform. This transform is very suitable to represent the noisy curved features that often exist in medicine or nano/micro images.

Tablet PCs are a new generation of notebook computers which provide multimodal input options of pen, voice and keyboard. Recently, these portable and flexible Tablet PCs have attracted attention as a potential tool in academic environments. This paper reviews the current use of Tablet PCs in teaching computer science and software engineering courses, presenting lectures and papers, and creating peer-review comments. The paper also presents applications of Tablet PCs in teaching and research at our university, the University of Canberra, Australia. These applications include marking assignments and reports, and developing signature verification applications.

In this paper, we describe the use of Artificial Intelligent (AI) techniques to transfer the animation from one 3D face to a newly created face. The animation is produced with the use of the vector muscle model. With this muscle model, in order to generate natural facial expressions in a 3D face, muscles have to be placed in the correct positions in the face and other parameters of the muscles need to be assigned suitable values. This is a heavy human-involved and time-consuming procedure. Our approach replaces this process by a faster and easier one, facilitated by AI techniques, in which the animator only has to select the faces with the most natural expressions from his point of view. First, we use Radial Basis Function (RBF) networks to deform a source face model to represent a target face model using the specification of corresponding landmarks on the two face models. We introduce a novel method to specify and adjust landmarks on the target face model automatically. The landmark adjustment process is done by Genetic Algorithms (GAs). After all the landmarks have been placed in optimal positions, the RBF networks are used to deform the source face model as well as to transfer the muscles on the source face model to the deformed face model. Finally, Interactive Genetic Algorithms are used to refine the parameters of muscles in order to produce high quality facial expressions.

Security administrators use network intrusion detection systems (NID systems) as a tool for detecting attacks and misuse, using passive monitoring techniques. However, there are sophisticated attacks which use ambiguities in protocol specifications to subvert detection. In these attacks, the destination endpoint reconstructs a malicious interpretation, whereas the passive NID system's protocol stack interprets the protocol as a benign exchange. There is a dire need for a new software element at the entry point of the network, which transparently modifies network traffic, so as to remove all possible ambiguities. This will ensure that all internal hosts and the NIDS interpret the traffic in a uniform way, hence removing all chances of an attack sneaking past the NIDS, unnoticed and unmonitored. In this paper, we will present the design and implementation of a normalizer whose job is to eliminate evasion and insertion attacks against an NIDS at the transport and network layers.

This paper proposes a general low cost framework to enable remote monitoring and controlling tasks in different environments, which leverage the strengths of uIP and web service technology. In this work, we further extend capabilities of uIP, the core tool to realize web services for embedded devices, to define a new standard for electronic devices to communicate using web services. The web service is provided as API to support three key operations with electronic devices including configuration, monitoring status and real-time controlling. This offers great flexibility and reusability as operations can be done remotely through the internet or mobile devices. In addition, the framework can be deployed fully or partially depending on the scale of target environments.

Migration of employees has posed a big problem for SMEs. Migration has many advantages but leads to difficulties for SMEs, too. One disadvantage caused by the migration of employees is that experience and knowledge lose when experienced employees leave. Training new employees consumes an enormous cost [1]. This paper discusses the problem how to compile the knowledge and reuse it to help mobile employees. The ontology methodology is chosen as a tool for modeling and storing knowledge. Ontology-based assistance is based on the work context, which employees are working on. Case-based reasoning technique is used to identify a suitable solution. The case study is a company that produces special ceramics for industrial applications. The project is developed in cooperation with experts from the company. The feedback from employees during implementation and exploitation in the company is the key factor for the project's success.

This paper proposes a new approach to estimate software costs using case-based reasoning. In this approach, the costs of a new software project are estimated by firstly retrieving the similar previous project and then adapting its costs to the current conditions. The approach focus on the estimation within a narrow context. The project is described as an ontology which allows the managers to estimate with various level of requirement analysis during the development. Moreover, the statistical analysis of previous works (i.e. the COCOMO model) are utilized to reflect the software development domain knowledge.

Process Pattern is an emerging approach to reuse process knowledge. This concept, though attractive, has not been well understood and exploited yet due to the lack of formalization and supporting methodology. In this paper, we present our work towards a more efficient application of process patterns. Our goal is making process patterns directly applicable in process modeling to facilitate process reuse. In order to achieve this objective, we firstly define a process meta-model integrating the process pattern concept. This meta-model provides a formal definition of process patterns and allows describing processes based on process patterns. Then we propose a method for building process models from process patterns. This method itself is described as a pattern-based process which can be tailored and automated to guide process designers in their work.

Currently, SMEs have to face a number of difficulties to survive in the market. Planning plays a crucial role for the management and control. SMEs operate on a dynamically changing environment, with many unexpected events. SMEs are differently organized; involve many complicated types of relationships between their basic entities. These factors lead to choose the MAS methodology as tool for modeling and controlling the running of SMEs. Each agent works on behalf of one business entity, and they are cooperative together to build the enterprise model. An ontology is chosen as a tool for representation of knowledge about enterprises. An ontology is used to form and store information describing an enterprise. It is a part of MAS and used for sharing information between agents. With the purpose to increase the robustness of enterprises, advance planning and scheduling (APS) is proposed as a methodology for solving planning problems. The research project is carried out in cooperation with an enterprise specialising in ceramic production for industrial applications.

To encourage a new novel generation of OLAP(On Line Analytical Processing) tools that enable users interact with multidimensional data in creative, visual, flexible and interactive way, there should be a sophisticated multidimensional data model (MDM) to represent natural hierarchical relationships of data related user domain of discourse. Furthermore, the ontology integration in this data model will leverages the interoperatibility among heterogeneous data warehouses. In this context, this paper introduces ontology-based MDM as an extended data model of MetaCube[1]. This model covers most of basic concepts, relationships of former model and provides a framework and mechanism to employ ontology in modeling process. Furthermore, this paper also gives an implementable OWL representation (a markup language of many of-the-shelf ontologies on the Web) for specification and data of this model.