摘要
人耳检测是人耳识别系统的第一个环节.在比较已有的人耳检测方法的基础上,介绍了一种复杂背景下的快速人耳检测与跟踪的方法.该方法主要分为两个阶段,离线级联分类器训练阶段和在线检测阶段.在离线训练阶段,首先结合人耳轮廓清晰,凹凸有致的特点,采用扩充后的haar-like型特征,依最近邻法则构造出弱分类器空间,然后根据经验选择GAB算法训练出强分类器,最后将多个强分类器级联成多层人耳检测器.在线检测阶段,为提高检测率,本文采用了调整分类器阈值和缩放检测子窗口的策略.最终检测器在CAS-PEAL人脸库上测试,检测率达到98%以上;在PⅣ1.7GHz的PC上对普通CMOS摄像头输入的320×240dpi视频进行人耳跟踪,速度可达6～7fps.实验结果表明,本文的人耳检测方法具有较好的实时性和一定的鲁棒性.
Ear detection is the first step of an ear recognition system. On the basis of comparing to other existing approaches, this paper introduces a kind of fast ear detecting and tracking approach under the complex background, which has two stages： off-line cascaded classifier training and on-line ear detection. In the stage of off-line training, considering the vivid contour, the concave and convex of the ear, we apply the extended haar-like features to construct the space of the weak classifiers using the nearest neighbor norms. And then we choose the GAB algorithm with experience to train the strong classifiers which form the cascaded multi-layer ear detector. In the stage of the on-line detecting, we adopt the methods of adjusting the threshold of the strong classifiers and zooming in out the detecting sub-windows for speeding up while keeping the size of the original image. The testing experiments on the face database of CAS-PEAL result an upwards of 98% hit rate, and on the 320 x 240 dpi video inputted by the CMOS camera using PⅣ 1.7GHz PC result a speed of 6 ~ 7fps, which shows that the proposed method is significantly efficient and robust.
出处
《中国图象图形学报》
CSCD
北大核心
2007年第2期222-227,共6页
Journal of Image and Graphics
基金
国家自然科学基金项目(60573058
60375002)
北京市教委重点学科共建项目(XK100080431)
