基于区域特征的视频序列人脸快速检测  

A Fast Face Detection Algorithm Based on Regional Feature for Video Sequences

作　　者：吴禄慎[1] 吴培敏[1] 

机构地区：[1]南昌大学机电工程学院,江西南昌330031

出　　处：《微电子学与计算机》2011年第5期162-165,共4页Microelectronics & Computer

基　　金：国家自然科学基金项目(59965003)

摘　　要：提出了一种基于区域特征的快速人脸检测算法.采用瞬时差分和背景差分获取并跟踪运动目标,消除了运动目标引起的背景模型更新误差.在检测到的运动目标区域内,通过基于区域特征的马赛克三分图模型检测人脸区域,并利用频率直方图方法合并所检测区域,最终获得人脸位置.实验结果表明,平均检测时间为30 ms/帧,检测准确率可达95.7%,算法复杂度低、检测效果好,适合各类视频图像的人脸实时检测.According to face detection for videos,a fast detection algorithm based on regional features was proposed in the paper.The comprehensive algorithm with temporal difference and background difference was used for acquiring and tracking the moving objects,which can avoid updating error of background model caused by moving object.In the region of moving objects,first Trisection image model of mosaic based on regional feature was used for detecting face region,then the frequency histogram method was adopted to combine detection area,and finally the face position was obtained.The experiment shows that the average detection time is 30 ms/frame and the accuracy rate obtains 95.7%.Therefore the algorithm which has lower complicacy and higher accuracy rate is suitable for real-time face detection for all kinds of video images.

关 键 词：马赛克规则 背景差分 瞬时差分 频率直方图 积分图 

分 类 号：TP317.4[自动化与计算机技术—计算机软件与理论]