基于新Haar-like特征的Adaboost人脸检测算法

为解决基于Haar-like特征的Adaboost人脸检测方法存在的特征计算复杂度较高的问题,提出两组Haar-like特征扩展集;利用积分图给出特征组的计算方法;采用Adaboost算法在正脸和侧脸样本库分别训练出正脸和侧脸级联分类器,并将其组成双通道分类器。在开源视觉库OpenCV上的实验结果表明,本方法具有较少的弱分类器数,检测效率高、计算速度快,对于多角度人脸检测具有较好的鲁棒性。

An improved adaboost algorithm based on new Haar-like feature for face detection

To solve problem of highly complexity of multi-angle face detection by the Adaboost algorithm based on the Haar-like, two new groups of extended Haar-like feature were proposed and the calculation method were exploited by the integral image. Then, the frontal faces cascaded classifier and the profile faces cascaded classifier was trained on the face database by the Adaboost algorithm respectively. Finally, the two-channel cascaded classifier was built. On OpenCV which is an open source vision database, the experimental results showed that the proposed method had better performance both in accuracy and computing speed, and could detect face with less weak classifiers. Meanwhile, the cascaded classifier had a good ability of robustness on detecting multi-angle face.