人脸特征检测与跟踪算法研究

【作者】 杨娟；

【导师】 茅耀斌； 王执铨；

【作者基本信息】 南京理工大学 ， 控制理论与控制工程， 2005， 硕士

【摘要】 人脸特征检测与跟踪作为人脸信息处理中的一项关键技术,近年来成为模式识别与计算机视觉领域一项受到普遍重视,研究十分活跃的课题。 本文首先对人脸特征检测和跟踪技术进行综述。在此基础上,分别研究了两种人脸特征检测和两种人脸特征跟踪算法。在人脸特征检测方面,第一种是基于先验知识和边缘提取的非模板人脸特征检测方法,该算法充分利用人脸及其特征的己知信息,通过形态学处理、面积滤波等操作得到眼睛和嘴唇的粗略位置,再利用边缘提取得到眼角和嘴角的准确位置;第二种是基于模板的人脸特征检测方法,即先利用基于颜色的方法检测到图像中的人脸候选区域,再通过眼睛检测来判断每个区域内是否存在脸,而眼睛检测则采用有效的模型匹配方法来实现。对于特征跟踪,第一种方法采用简单的模板匹配方式,即利用上一帧特征区域作为模板,来搜索匹配当前帧中的特征,再进行边缘提取得到眼角和嘴角的位置;另一种是基于变形模板和SNAKE的人脸特征跟踪方法,先用变形模板确定特征轮廓的大致位置,再利用SNAKE方法自主收敛到轮廓的准确位置,在此基础上用双SNAKE实现唇动跟踪。 通过实验,证明了以上算法的可行性。同时,比较分析了上述算法的利弊。最后,给出进一步的研究方向。更多还原

【Abstract】 The problem of facial feature detection and tracking, a key technology in face information processing, has recently emerged into an important subject in the fields of pattern recognition and computer vision.Based on a general review of existing schemes on this particular topic, we discuss two facial feature detection algorithms and two tracking algorithms. The first detection method is based on facial knowledge and edge extracting, which localizes eyes and mouse roughly using morphologic processing and area filter, and then extracts canthi and corners of the mouth accurately using Canny edge extracting method. The other detection algorithm is based on an eye model, which first uses a statistical skin-color model to segment face-candidate regions in the image, and the presence or absence of a face in each region is verified by means of an eye detector, based on an efficient template matching scheme. For the two facial feature tracking algorithms, the former approach uses simple template matching to search eyes and mouse in the current frame. The second tracking method synthesizes deformable template and snake, which localizes feature contours using deformable template, then convergences to the accurate feature contours using snake. Later, we discuss mouse tracking based on double snakes.Finally, several further research aspects are pointed out based on the analysis of the cons and pros of the above four algorithms更多还原