一种基于偏转角度的AAM人脸特征点定位方法

摘要

For face feature localization that has deflection angle, model angle and the initial position of fitting have a great effect on face feature localization. However, traditional AAM (Active Appearance Models) face feature localization method do not think about this problem, the location accuracy and speed for the face with deflection angle feature points is not ideal. A method based on deflection angle was proposed in mis paper to solve this problem, a kind of method is put forward. This method uses two eyes and mouth center coordinates to calculate face deflection angle and find the initial position of fitting and templates. Adaboost algorithm and facial skin properties in YCbCr color space are applied to pre-detection of facial features in the images, according to the features area that is found in pre -detection the deflection angle was calculated, inverse compositional algorithm combined with the template of the angle were used to locate feature points. Experimental result shows that for feature points localization of face which have deflection angle, this method are improved than traditional method in both accuracy and speed.%对有偏转角度的人脸特征点定位来说,拟合初始位置和模型的角度对人脸特征点定位效果有很大的影响.而传统的AAM(Active Appearance Models)人脸特征定位方法没有具体考虑这一问题,对有偏转角度的人脸特征点的定位准确率和速度并不理想.为解决这个问题,文中提出了一种利用两眼中心坐标和嘴中心坐标来计算人脸偏转角度,根据坐标和角度确定拟合初始位置和模板的方法.用Adaboost和YCbCr对人脸进行预检测,根据找到的特征区域计算偏转角,用反向算法结合该角度的模板进行特征点定位.实验的测试结果表明本方法对有偏转角度的人脸的特征点定位比传统方法在准确度和速度上都有了提高.