Robust lip tracking using active shape models and gradient vector flow

Abstract

Lip movements and configurations provide useful information which can be utilized to improve automatic speech and speaker recognition. However the use of this visual information requires accurate lip tracking algorithms. A new technique is outlined that is able to estimate the outer lip contour under noisy conditions accurately. It is based on the combination of gradient vector flow (GVF) force fields and active shape models (ASM). Results are presented for speech reading using a speaker dependent database. The effectiveness of this algorithm on noisy edge maps generated from chromatic segmentation is also demonstrated.