摘要
提出一种新的非线性变换的彩色空间 YC″r C″b,利用次高斯概率分布函数拟合皮肤色度信息 ,得到候选区域。为了排除候选区域中的非人脸 ,首先根据均值和方差信息分割出候选区域中的纹理特征信息 ,再通过多尺度形态边缘检测算子检测候选区域的边缘 ,利用 PCA边缘方向 ( PCAED)信息定位眼睛 ,然后根据人脸特征的几何形状信息定位其他特征 (鼻、嘴 ) ,通过这些几何特征信息对肤色分割得到的候选区域进行验证 ,最终得到正确的人脸区域。利用 3个实验数据集测试该算法 ,并与其它相应的算法相比较 ,提出的非线性彩色空间对于肤色分割具有很好的效果 ,且对光照和姿态具有良好的不变性。另外 ,利用 PCAED信息和几何特征信息检测人脸特征具有很高的定位精度 ,定位检测率优于其他方法。实验结果表明 ,该算法具有定位准确率高 ,漏检率和误检率低等特点。
A novel approach for skin segmentation and facial feature extraction is proposed. The proposed skin segmentation is a method for integrating the chrominance components of nonlinear YC″ rC″ b color model. The chrominance components of nonlinear YC″ rC″ b color space are modeled using a subgaussian probability density function,and then the face skin is segmented based on this function. In order to authenticate the face candidate regions,firstly texture information in face candidate regions is segmented using mean and variance of luminance information,and then the eye is located by the PCA edge direction information. Finally,the others features,such as nose and mouth,also are detected using the geometrical shape information. As all the above-mentioned techniques are simple and efficient,the skin segmentation based on nonlinear color space method has the invariability of lighting and pose. In the experiments,the method has been successfully evaluated using three different test datasets. Experimental results indicate that the method has a high detection ratio, a low false detection and a low missing detection ratio.
出处
《数据采集与处理》
CSCD
2004年第2期160-166,共7页
Journal of Data Acquisition and Processing
关键词
人脸检测
人脸自动识别
非线性彩色空间
边缘检测
face detection
nonlinear color space transform
PCA edge direction (PCAED)
eye location
facial feature
