Accurate and real-time lip contour extraction based on constrained contour growing

Abstract

In this paper, a novel method for accurate and real-time lip contour extraction in gray-level images is presented. The main idea is to grow the contour from some key points based on constraints of parabola fitting and gradient maximum which are combined using D-S theory. This approach can accurately extract continuous lip contour by simultaneously taking color and shape information into consideration, and achieve real-time performance using less computation without iteration and modeling, which can provide high efficiency to meet the need of real-time application. Furthermore, this method is applied in a real-time lipreading system using visual features constructed from lip contour information extracted by the proposed method. Experimental results show that both the lip contour extraction and lipreading system achieved good performance, which demonstrates the effectiveness and efficiency of the proposed method.