This paper describes a real time face detection and tracking system. The method consists in modelling the skin face by a pixel fusion process of three colour sources within the framework of the Demster-Shafer theory. The algorithm is composed of two phases. In a simple and fast initialising stage, the user selects successively in an image, a shadowy, an overexposed and a zone of mean intensity of the face. Then the fusion process models the face skin colour. Next, on the video sequence, a tracking phase uses the key idea that the face exterior edges are well approximated as an ellipse including the skin colour blob resulting from the fusion process. As ellipse detection gets easily disturbed in cluttered environments by edges caused by non-face objects, a simple and fast efficient least squares method for ellipse fitting is used. The ellipse parameters are taken into account by a stochastic algorithm using a particle filter in order to realise a robust face tracking in position, size and pose. The originality of the method consists in modelling the face skin by a pixel fusion process of three independant cognitive colour sources. Moreover, mass sets are determined from a priori models taking into account contextual variables specific to the face under study. Hence, the face specificity which is to present shadowy (neck) and overexposed zones (nose, front) is considered, so that sensitivity to lighting conditions decreases. Results of face skin modelling, fusion, ellipse fitting and tracking are illustrated and discussed in this paper. The limits of the method and future work are also commented in conclusion

This article deals with facial segmentation and liptracking with feedback control by face model synthesis. On this topic, the search community is divided into two parts : analysis and synthesis. We want to use all the knowledge to create a global analysis/synthesis chain where the image analysis needs the 3D synthesis and conversely. As it happens, applications like face tracking or augmented reality need a rapid, robust and descriptive-enough solution. Our solution is based on a two step approach: the first step is a real-time facial segmentation with active contour models and the second step recovers a 3D-face model in order to extract more precise parameters to adjust the first step. The contribution of this paper is to couple two research fields for creating a real time application. The results obtained show rapid and robust performances which could be exploited in a more global real-time face tracking application

Handling the inter-flow synchronization (e.g. images and sound of a video) in distributed multimedia applications seems to be fundamental. In this paper, we propose a new component model, named OSAGAIA , in order to tackle synchronization between multimedia flows from the source (e.g. media capture) to the destination (e.g. media player). This model is made of two entities. The first one is called "elementary processor" and is used as a runtime environment (container) for multimedia components. The second one is called "conduit" and is used to transport synchronous multimedia flows within the application. To ensure the synchronization within both entities, we sample multimedia flows in order to constitute temporal units and synchronous slices. We test this mechanism with a distributed prototype implemented with Java language using JMF (Java media framework) API and TCP/IP as the network protocol. This prototype shows that synchronization is kept between a non-processed and a processed image flow.

For a video surveillance application based on wireless transmission of images acquired with a static camera, we investigate two features within JPEG2000, namely the choice of a nonlinear color transform in order to improve color rendering at very high compression ratios, and the unsupervised extraction of regions of interest (ROI) by motion detection.

This paper deals with the low-level joint processing of color and motion for robust face analysis within a feature-based approach. To gain robustness and contrast under unsupervised viewing conditions, a nonlinear color transform relevant for hue segmentation is derived from a logarithmic model. A hierarchical segmentation scheme is based on Markov random field modeling, that combines hue and motion detection within a spatiotemporal neighborhood. Relevant face regions are segmented without parameter tuning. The accuracy of the label fields enables not only face detection and tracking but also geometrical measurements on facial feature edges, such as lips or eyes. Results are shown both on typical test sequences and on various sequences acquired from micro- or mobile cameras. The efficiency of the method makes it suitable for real-time applications aiming at audiovisual communication in unsupervised environments.

Previously, we developed a method and a distributed platform for the re-engineering of applications by adding cooperation. The goal was to supply a way of communication based on the exchange of events, messages and shared data. Now, we propose to adapt this approach (method and platform) to distributed multimedia applications. These applications present the characteristic to be organized around the communication. So, we can consider them as composed of distributed components which have to collaborate. Since it is heavily interactive, such an application needs to adapt itself in real-time to the user and to the environment in which it runs. Our approach consists of breaking down the application into two levels: the first one reflects the users' point of view in terms of functionalities; and the second one reflects the way to achieve these functionalities in terms of quality of service. We propose to organize components into workgroups and subworkgroups corresponding to this two levels. The platform manages these groups and makes them evolve in real time. It also ensures the interoperability of components which cooperate in these groups.

An unsupervised algorithm for speaker's lip segmentation is presented. A color video sequence of the speaker's face is acquired, under natural lighting conditions and without any particular make-up. First, a logarithmic color transform is performed from the RGB to HI (hue, intensity) color space and sequence dependant parameters are evaluated. Second, a statistical approach using Markov random field modeling segment the mouth shape using the red hue predominant region and motion in a spatiotemporal neighborhood. Simultaneously, a region of interest (ROI) is automatically extracted. Third, the speaker's lip shape is extracted from the final hue field with good quality results in this challenging situation.

This paper describes the real time implementation of a simple and robust motion detection algorithm based on Markov random field (MRF) modeling, MRF-based algorithms often require a significant amount of computations. The intrinsic parallel property of MRF modeling has led most of implementations toward parallel machines and neural networks, but none of these approaches offers an efficient solution for real-world (i.e., industrial) applications. Here, an alternative implementation for the problem at hand is presented yielding a complete, efficient and autonomous real-time system for motion detection. This system is based on a hybrid architecture, associating pipeline modules with one asynchronous module to perform the whole process, from video acquisition to moving object masks visualization. A board prototype is presented and a processing rate of 15 images/s is achieved, showing the validity of the approach.