[Show abstract][Hide abstract] ABSTRACT: An algorithm for speaker's lip contour extraction is pre- sented in this paper. A color video sequence of speaker's face is acquired, under natural lighting conditions and wit h- out any particular make-up. First, a logarithmic color transform is performed from RGB to HI (hue, intensity) color space. A bayesian approach segments the mouth area using Markov random field modelling. Motion is combined with red hue lip information into a spatiotemporal neigh- bourhood. Simultaneously, a Region Of Interest and rele- vant boundaries points are automatically extracted. Next, an active contour using spatially varying coefficients is ini- tialised with the results of the preprocessing stage. Final ly, an accurate lip shape with inner and outer borders is ob- tained with good quality results in this challenging situa- tion.
[Show abstract][Hide abstract] ABSTRACT: Active contours or snakes are widely used in object segmentation for their ability to integrate feature extraction and pixel candidate linking in a single energy minimizing process. But the sensitivity to parameters values and initialization is also a widely known problem. The performance of snakes can be enhanced by better initialization close to the desired solution. We present a fine mouth region of interest (ROI) extraction using gray level image and corresponding gradient information. We link this technique with an original snake method. The automatic snakes use spatially varying coefficients to remain along its evolution in a mouth-like shape. Our experimentations on a large image database prove its robustness regarding speakers change of the ROI mouth extraction and automatic snakes algorithms. The main application of our algorithms is video-conferencing
[Show abstract][Hide abstract] ABSTRACT: In this paper a coarse-grain bus architecture, which provides flexible and reconfigurable solutions for vision machine development is presented. This bus architecture is host-independent and allows the use of heterogeneous application boards. These hardware modules are linked to a fast bus, called a DRIFT bus, via a standard interface which we define here. The DRIFT bus is a dynamically reconfigurable bus, meaning that data paths can be changed during the process to fit the communication requirements. It allows high-speed data block transfers at 1 Gbit/s. Moreover, it supports different transfer modes, such as parallel and pipeline transfers. To validate our bus concept a prototype has been developed. We give the most important performances obtained with this machine and some examples of real-time video algorithms (Difference Of Frames, Multi-pass filtering) are presented. Being based on a fast and flexible bus, our bus architecture can be used in wide-range vision applications.
No preview · Article · Jun 1998 · Real-Time Imaging
[Show abstract][Hide abstract] ABSTRACT: Different existing vision machines are surveyed. The authors are
particularly interested in their intercommunication system, which is a
fundamental factor in the overall performance. In order to fulfil the
need for a flexible, high-performance, and cost-effective vision machine
a coarse-grain architecture is proposed. It is based on a dynamically
reconfigurable fast bus. The authors are developing a prototype machine.
This prototype, hosted by a PC, uses a VME-like back plane cabinet in
which the P1 connector is used for the configuration control lines and
the P2 connector for the implementation of 4 fast transfer lines
channels. The DRIFT interface is designed to achieve a bandwidth of 1
Gbits/s. Processing modules are built on extended double Eurocards on
which the DRIFT interface occupies 25% of the whole board surface. This
prototype will be used in some vision applications in which broadcasting
and dynamic reconfiguration are necessary