eNTERFACE'11 invited speakers

Nick Campbell

See program page.

Talk: Multimodal processing of spoken interaction - managing contact
This talk describes some recent work in interactive speech synthesis technology. It is no longer enough for a machine simply to be able to talk; it must now be able to detect the reactions of the listener in order to be able to adapt its speech output accordingly. With this goal in mind, we have been testing a small robot dialogue device that detects the presence of people in its environment and starts a conversation with them.
The conversations are all recorded in audio and video and the goal of the robot is to obtain a signed consent form from each dialogue participant so that their data may be used as part of our research. The robot was exhibited in a public science gallery and we obtained more than a thousand signed consent forms over a period of three months. The present talk discusses the issue of managing "contact" in such a dialogue, and proposes methods for detecting not just the presence of people, but also their interest in the conversation.

Václav Hlaváč

See program page for dates.

Talk 1: Milestones of image analysis and computer vision
Slides: Milestones of image analysis and computer vision (3MB)
This introductory lecture to aims at explaining the image processing, analysis and computer vision disciplines from their development-in-time perspective. The concepts, tasks formulations and principal contributions will be mentioned with links to principal contributions. The views of signal processing, image formation physics, human image perception, image processing, image analysis, mathematical morphology, 3D computer vision, image sequences understanding, etc. will be explicated. The links to neighboring disciplines of pattern recognition, machine learning, artificial intelligence, and computer graphics will be mentioned too. The broad view of the discipline, in the presenter's opinion, may help the people entering the field to avoid some mistakes which many of us made before them.

Talk 2: Structure in computer vision and pattern recognition: Why doesn't it really fly high?
Slides: Structure in computer vision and pattern recognition: Why doesn't it really fly high? (66MB)
Structural pattern recognition has been highly popular in 1960-1970s with seminal contributions from Kung Su Fu. The later interest in it faded with the rise of statistical approaches which penetrated even mathematical linguistics where structural analysis was originally started by Noam Chomsky in 1950s. With images, the failure of structural method was glaring.
Recently the structural methods have been mostly understood as the graph embedding (H. Bunke's K.S. Fu Award lecture at ICPR 2010).
In the talk, I will talk about my view of the subject originating from methods from our book Schlesinger M.I., Hlavac V.: Ten lectures on statistical and structural pattern recognition, Kluwer 2002. The structure is also needed for a large scale outdoor scene analysis. I will explain the topic and relate it to my own work and the work of my collaborators: (1) Repetitive structures in house facades, (2) in using structure (pose primitives) in analysis of human activity from video; (3) in using 2D context-free grammars for analysis of mathematical formulae, etc.