Multiobjectives genetic snakes: application on audio-visual speech recognition

In this article a new optimization of genetic snakes (GS): multiobjectives genetics snakes (MGS) faster and simpler to implement is proposed. They enable us to make converge two snakes in parallel while minimizing energies of different nature. We apply them to the modeling of mouth contours within the framework of the audio-visual speech recognition (AVSR). The proposed AVSR system implements a classifier based on the STM (sparse template matching) which simplicity makes it possible to consider a real time implementation. We evaluate the classifier performances on European database M2VTS, and compare the performances of the GS and MGS.