非监督、多级嘴唇区域分割方法

Unsupervised and Hierarchical Lip Segmentation Method

该文介绍了一个非监督、多级嘴唇区域分割检测方法。首先提出利用fisher变换增强嘴唇区域,然后利用嘴唇在人脸区域的分布面积比,提出利用统计阈值完成嘴唇的初步分割,最后利用嘴唇对称性和轮廓光滑性的先验知识,提出了基于局部阈值调整完成嘴唇的精细分割。实验证明该方法在不同光照条件下、对不同人、各种表情都能自动地、鲁棒地、精确地检测出嘴唇区域,并较准确提取初步的嘴唇几何参数。利用这些几何参数作为轮廓定位的初始化条件,可以大大提高变形模板和ASM模型等嘴唇轮廓定位算法的速度和准确度。

This paper proposes a novel unsupervised and hierarchical method to segment lip region.In this method,Fish-er transform is first employed to enhance lip region in the face image.Secondly,lip region is coarsely segmented by us-ing a statistical threshold learnt from the distribution statistic of the ratio of the lip area to the whole face region.Final-ly,the authors employ the prior knowledge of the lip symmetry and smoothness to adjust the local adaptive thresh old so as to refine the segmentation.Experiments show that our method can segm ent lip region automatically,robustly,and accu-rately,and it is insensitive to illumination variance,identity and expression changes.Some geometric feat ures can be ex-tracted simultaneously.With the help of these initial measures ,the performance of the lip contour locating methods such as deformable templa te and ASM(Active Shape Model)is promoted dramatically.