一种快速、鲁棒的唇动检测与定位方法  

An Approach to Robust and Fast Locating of Lip Motion

作　　者：王瑞[1] 高文[1] 马继涌[2] 

机构地区：[1]哈尔滨工业大学计算机科学与工程系,哈尔滨150001 [2]中国科学院计算技术研究所,北京100080

出　　处：《计算机学报》2001年第8期866-871,共6页Chinese Journal of Computers

基　　金：国家自然科学基金 (6 978930 1);中国科学院百人工程;国家"八六三"高技术研究发展计划"多功能感知机"项目基金资助

摘　　要：提出了一种利用唇色在彩色空间分布信息和唇在人脸区域的几何分布特性进行快速、鲁棒的唇动检测与定位的方法 .在该方法中 ,首先提出采用有约束的 Fisher变换进行嘴唇图像增强 ;然后利用嘴唇在人脸区域的几何分布特性 ,来提高唇动定位的速度和准确性 ,唇动定位主要利用两种几何分布特性 :(1)利用唇的区域面积和人脸的区域面积比值变化范围不大这一特性自适应确定区分唇色和肤色的阈值 .(2 )利用外唇参数与内唇参数之间有近似线性关系的假设 ,用外唇参数预测内唇参数 ,克服内唇定位经常陷于局部最小、定位不准的缺陷 .实验证明 ,该唇动检测与定位系统无论从速度还是从可靠性、准确性都能满足应用的要求 .The problem of fast and robust lip locating becomes more and more important in multimode interfaces such as lip reading, facial expression recognition, lip motion synthesis and Synthetic/Natural Hybrid Coding (SNHC) in MPEG4. However, locating and tracking lip motion accurately are very difficult and challenging because the lip contours of different people vary greatly and are affected by lip variations, head movement and illuminant conditions while a speaker is uttering. A fast and robust lip detecting and locating approach which combines the information of lip color in color space and the geometrical feature of lip color in human face space is proposed in this paper. First, the Fisher transform with constraints is proposed to enhance the lip image. Then, the geometrical feature of lips in human face is proposed to increase the accuracy and speed of lip locating. Two key characteristics were found. One is that the ratio of lip color area to that of skin color is almost unchanged. This characteristic is used to adaptively set the threshold to separate the lip color and skin color. The other is the linear relation between parameters of inner lips and parameters of outer lips. The parameters of inner lips can be predicted by parameters of outer lips to overcome the local minimum problem of locating inner lips. The experimental results have shown that the approach to locate lips is very efficient both in locating speed and locating accuracy.

关 键 词：FISHER变换 多元线性回归分析 唇动检测 图像增强 定位 计算机 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]