作者: Meng Li , Yiu-ming Cheung
DOI: 10.1007/978-3-642-15819-3_51
关键词:
摘要: This paper addresses the problem of lip segmentation in color space, which is a crucial issue to success lip-reading system. We present new approach contour extraction by taking account difference between and skin space. Firstly, we obtain segment sample via transformation sequence 1976 CIELAB LUX spaces. Secondly, establish Gaussian model make use hue saturation value each pixel within estimate parameters. Subsequently, memberships non-lip regions are calculated, respectively. Thirdly, employ morphological filters desirable region approximately based on memberships. Finally, extract convex hull algorithm with prior knowledge human mouth shape. Experiments show efficacy proposed comparison existing methods.