摘要
根据由粗到精的思路,综合利用了眼睛器官的特征、人脸模板以及基于人脸区域象素分布的统计信息等线索.首先用基于AdaBoost学习算法的级联模型发现可疑人脸区域,然后在可疑人脸区域内发现可疑眼睛区域并组合可疑眼睛区域对得到候选人脸区域,最后用人脸模板验证候选人脸区域,精化人脸定位.实验表明,本算法能快速而精确地实现多人脸检测与定位,且对彩色图像和灰度图像都适用.
The algorithm proposed in this paper following the thought of “coarse-to-fine”, makes use of features of eyes, face templates, statistical information based on the distribution of pixels in the face region and so on. First, suspicious regions that may contain faces are discovered by a cascaded model based on AdaBoost learning algorithm. Second, eye-analogue segments are discovered in the suspicious region and then a pair of eye-analogue segments is treated as left and right eyes in a face pattern if their placement is consistent with the anthropological characteristic of human eyes. The image patch containing these segments is selected as a face candidate. Third, accurate face regions are obtained by eliminating false positives which can not pass the verifier composed of face templates. Experimental results show that this algorithm can detect and locate multifaces accurately and rapidly, suitable for color or gray level images.
出处
《小型微型计算机系统》
CSCD
北大核心
2005年第9期1520-1524,共5页
Journal of Chinese Computer Systems
基金
江苏省"十五"科技攻关计划(BE2001028)资助
江苏省自然科学基金(BK2004079)资助.
