TUTORIALS PROGRAMME
26th September 2006
| SESSION A: COGNITIVE METHODS |
| |
| 14:00-15:30 |
The semantic gap in image retrieval
Prof. Ebroul Izquierdo, Queen Mary College London (University of London) |
| |
Motivation:
The fast development of innovative tools to create user friendly and effective multimedia libraries, services and environments requires novel concepts to support storage, annotation and retrieval of huge amounts of digital audiovisual data. The availability of a wide range of digital recorders accessible to anyone, from cheap digital cameras to complex professional movie capturing devices, is making possible that the stocks of digital libraries already packed with visual content continue rising rapidly. However, the lack of automatic indexing and retrieval technology is rendering useless significant portions of this data.
To get closer to the vision of useful multimedia systems, the indexing and search technology to be employed need to be efficient and user friendly adding value to content by making straightforward to find it when needed. Much work on image indexing and retrieval has been conducted over the past few years. On the one hand research has focused on the definition of low-level descriptors and the generation of metrics in the descriptor space. On the other hand, significant work has been also carried out in the generation of conceptual relations and ontologies for general and specific application domains. Although content-based low-level descriptors are extremely useful in Content Based Image Retrieval systems, they have little in common with high-level semantic concepts. Consequently, the challenge remains the same: how to link semantic query structures which appear natural to humans with low-level content based descriptors in order to bridge "The Semantic Gap" in multimedia processing. Here the Semantic Gap is defined as:
“the large disparity between the low-level features or content descriptors that can be computed automatically by current machines and algorithms, and the richness and subjectivity of semantics in user queries and high-level human interpretations of audiovisual media”. |
| |
Aims and content:
The main goal of this tutorial is to present the latest developments at both sides of the semantic gap. Current trends towards the linking of content based analysis and semantics will be discussed. Application scenarios will be presented in which technology for semiautomatic semantic annotation and retrieval has been successful to some extend.
The vision of integrated knowledge, semantics and content engineering for narrowing the semantic gap will be outlined.
Learning outcomes:
- Understanding of fundamental techniques to structure and annotate visual content
- Understanding the main research issues related to the semantic gap and applications
- Understand current trends in related research
Content Description
- Content Based image retrieval
- Metadata, descriptors and descriptors spaces
- Similarity metrics for visual content
- Semantic based annotation
- Relevance feed-back and multi-descriptors spaces
|
| |
| Back to Top |
| |
| 16:00-17:30 |
Cognitive Vision techniques for Video Analysis and Understanding
Dr Monique Thonnat, Director ORION Team, INRIA (Sophia Antipolis, France) |
| |
Motivation:
There is an increasing demand in real-time robust video analysis systems. This is particularly true for visual surveillance applications where the number of video cameras is growing faster than the number of security operators. Intelligent video understanding systems can filter the information coming from the video cameras by detecting interesting behaviours. Recently huge progresses have been made both in the performances of the hardware (computing power and image quality) and more importantly in the design of video understanding systems. We focus here on the second point how to build efficient and effective intelligent video understanding systems. It implies to solve two problems: first how to integrate in a real-time system and to control a set of video analysis techniques and second how to recognise video events related to the behaviour of physical objects in the real world.
Intended Audience:
Students, research scientists and engineers working in computer vision, in video analysis or in video understanding. |
| |
Aims and Objectives:
The objective of this tutorial is to describe the state of the art for integration and control of video analysis techniques and for video understanding. Examples will be shown for video surveillance applications. The emphasis will be put on knowledge representation and reasoning rather than on low level vision techniques.
Contents:
- Introduction to programme supervision techniques
- Knowledge representation of video analysis operators
- Planning and control of execution of video operators
- Introduction to video understanding
- Knowledge representation of video events
- Recognition of video events
|
| |
| Back to Top |
| |
| Session B: Image Processing Tools and Techniques |
| |
| 14:00-15:30 |
Level Set Function Inspired Curve Evolution and Related PDEs
Prof. Dipti Prasad Mukherjee, Computer and Communication Division, Indian Statistical Institute, Kolkata |
| |
Motivation
Evolution or deformation of a closed curve in an image can be implemented using evolution of a level set function of the closed curve. The curve evolution process is an established tool in the field of image processing, applications of which are found in the areas of image segmentation, tracking etc. If curve evolution is considered as a tracking of interface problem, then a suitable PDE can be designed, either as an initial value or a boundary value problem to evolve the level set function of the closed curve. Zero level set value of the evolving level set function is considered as the position of the evolved curve at different time instants. Understanding of this PDE and the design of a suitable level set function of the closed curve are prerequisites for identifying the applications of level set based curve evolution. The objective of this tutorial is to clearly understand the mechanics of PDE governing evolution of level set function, which, in turn, evolves the curve. The tutorial will also point to several emerging and interesting applications of level set based curve evolution, for example, curve interpolation, pattern generation or open-ended curve evolution. It will also discuss the edge of level set based curve evolution compared to the related existing image processing applications based on curve evolution.
The motivation for a tutorial comes from the fact that given the state of development in the field of level set analysis, there exists literature that explores the mathematical basis of level set analysis in great detail. On the other hand publications detailing the implementation aspects, including numerical issues have not yet been divulged to a satisfactory extent in the technical publications. The aim of proposing this tutorial is to bridge this gap so that far more image processing scientists can get to know the implementation aspects of otherwise tedious PDEs and can explore novel applications of level set analysis.
The session will explore techniques of level set analysis to suit different kinds of applications, like those related to image segmentation, tracking, evolution of open-ended curves, pattern generation, contour interpolation etc. Two of Prof. Mukherjee publications (IEEE Trans. Image Processing, April 2004 and June 2004) discuss applications of level set and modify the existing PDEs of the active contour paradigm; further, his publications involving (a) open-ended curve evolution and (b) pattern generation and contour interpolation, all using different forms of PDE of level sets, are currently being reviewed by different journals.
Intended Audience
The expected pre-requisite for the audiences is the basic knowledge of rudimentary image processing tasks (for example, some exposure to image segmentation, tracking, edge detection etc., which I believe a commonplace for VIE attendees). Presentations will start from a fairly basic level and then continue to explore insights from the introductory problems. Both researchers and engineers will benefit from the discussion of both theoretical and implementation issues. Final year undergraduate and graduate students intending to do projects or thesis can find material suitable for challenging projects using level set analysis. |
| |
Aims and Objectives
I believe that there exists a number of areas in the field of visual information engineering where a successful application of level set methods are long over due. I also believe that even though researchers are taking interest in this particular area, scope exists by which a larger section of researchers in the areas of image processing and computer vision can be encouraged to explore fundamentals of level set analysis. The bottleneck is good technical publications in this field where both theoretical and implementation aspects are harmonized for a wider audience. I believe that the proposed tutorial will be a step towards achieving this goal. The specific achievable goal that I can envisage is that:
(a) Increase awareness and understanding of the method of level set analysis
(b) Numerical implementation scheme of PDE governing evolution of level set function and its connection to curve evolution
(c) Exploring different application areas of level set analysis (for example, applications in graphics, applications in 3D shape analysis, applications in biomedical image processing etc.)
Contents:
(a) Why curve evolution? – Few slides on applications of curve evolution.
(b) How is curve evolution posed as a problem of evolution of level set function?
(c) Comparison of level set based curve evolution with existing curve evolution approaches.
(d) How can level set function be designed?
(e) How evolution of level set function can be stopped at the point of interest inside an image?
(f) Computational complexity and possible ways to minimize computational complexity.
(g) Applications of level set based curve evolution: (i) Image Segmentation (ii) Tracking single or multiple objects in video (iii) Shape interpolation (for example, interpolating shapes in between two frames in a video sequence) (iv) Clustering feature space (v) Pattern generation.
(h) Research issues: The curve evolution discussed so far is a closed contour – can this be extended for thin filament like open-ended curves.
|
| |
| Back to Top |
| |
| 16:00-17:30 |
Scalable Image Coding with JPEG2000
Shailesh Ramamurthy, Motorola India Electronics Bangalore |
| |
Motivation
JPEG2000 (ISO/IEC 15444), the new generation still image coding technology, offers a scalable, unified coding framework for a variety of still image types (bi-level, gray-scale, color, multi-component) with very different characteristics (natural, scientific, medical and remotely sensed images, text, rendered graphics, etc). Areas where this technology scores over existing still image coding standards include superior low bit-rate operation, progressive spatial as well as quality scalability, transmission in noisy environments, Region of Interest Coding, and, a unified system for lossy and lossless coding. JPEG2000 enables the paradigm of ‘Encode once, decode in many ways’ by way of its design and allowing for powerful random access features into the bits-stream. The encoder can, for instance, losslessly encode image data, while the decoding application can choose the resolution, component, position or quality that it is interested in. It is possible to extract and decode bytes required for an intended application without having to decode the entire bit-stream. JPEG2000 is a powerful vehicle for highly scalable compression enabling resolution, quality, position and component scalability, along with excellent coding performance in terms of enabling high quality at high compression ratio. Potentially, it has a large application base for a number of future imaging applications in diverse areas.
Intended Audience
This tutorial targets engineers, researchers and hardware designers interested in JPEG2000 technology. The participants are assumed to have familiarity with signal and image processing concepts. |
| |
Aims and Objectives
- Acquaint the participants with the important aspects of the core technology in JPEG2000
- Delineate the latest developments and directions taken by the Standard
Contents
This tutorial covers the algorithmic and implementation aspects of JPEG2000, and includes the following modules:
- Introduction to JPEG2000
- Data ordering and Code-Stream formation in JPEG2000
- The Wavelet Transform
- Bit-plane coding: The Coefficient Bit Modeler and Arithmetic Coder
- Rate Control in JPEG2000 – an overview
- Future directions
Apart from the theoretical aspects covered in this tutorial, the concepts will be illustrated with relevant examples and demonstrations. |
| |
| Back to Top |
| |
|
