摘要
针对面向唇读的水平集模型在嘴唇分割中存在边界过收敛和过早收敛的问题,文中提出了一种改进的基于先验知识的水平集模型(简称为PS-Level Set)来进行嘴唇几何形状的定位.PS-Level Set模型利用改进的差值能量函数引入嘴唇形状的先验信息.在曲线演化过程中,反复比较演化曲线和先验曲线的差距,使曲线的演化形状逐渐逼近先验模型形状,从而更精确地收敛于目标物体实际轮廓.实验表明,用PS-Level Set模型定位嘴唇几何形状的准确率比用水平集模型提高了8.38%.
In order to overcome the overconvergence and the premature convergence of lip boundary caused by the level set model for geometric shape detection, an improved level set model based on the prior shape （ PS-Level Set） is proposed. In this model, the prior shape information of lip is incorporated into an improved differential energy function, and the differences between the evolution shape curve and the prior shape curve are repeatedly compared during the curve-evolving process, which enables the evolution shape to gradually approach the prior one and to converge to the target object more accurately. Experimental results show that, as compared with the conventional level set model, the proposed model improves the detection accuracy by 8.38%.
出处
《华南理工大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2010年第2期121-125,共5页
Journal of South China University of Technology(Natural Science Edition)
基金
国家自然科学基金资助项目(60572141
60602014)
关键词
唇读
形状定位
水平集模型
曲线演化
lip reading
shape detection
level set model
curve evolution
