一种基于正反 Adaboost 的人脸检测算法

摘要

Accurately locating and tracking the position of the user's face is important for autostereoscopic display. A cascade face detector learnt by using real Adaboost algorithm is presented, and for the purpose of reducing the false alarm, a novel procedure called Frontal-Inverse Adaboost is proposed, then a process to gain accurate eyes' loca-tions is executed with minimum filter and integral image. A set of experiments are presented, the results indicate that the precision improving to normal Adaboost algorithm is satisfying. It can be seen that the proposed method is very efficient and has lots of significant values to autostereoscopic display.%人脸的检测和跟踪对于自由立体显示非常重要.本文采用连续 Adaboost 算法学习出一种瀑布型的人脸检测器,同时针对 adaboost 算法误检率比较高的弊病,提出了一种正反 Adaboost 检测的思想,最后通过利用已有积分图以及最小值滤波,完成了对人眼的识别操作.实验结果表明,算法对现有 adaboost 算法精度上的改进令人满意,基本能满足自由立体显示的需要.