提出一种基于沃尔什特征的增强型AdaBoost 人脸快速检测算法，不仅具有很快的训练速度，而且利 用较少的非人脸样本进行训练就可以达到较好的检测效果。首先，提出用较少的沃尔什特征来代替大量的 Harr-Like 特征可以较大幅度的降低特征之间的冗余。然后提出一种双阈值增强型AdaBoost 算法，其中双阈 值的快速搜索方法大大节约了训练时间，并且在训练Cascaded 检测器过程中，前层分类器的训练结果对后层 分类器的训练具有指导作用，加强了总体检测器的性能，另外通过各层分类器阈值的调节，能够将人脸和非 人脸的训练结果尽量分离。最后，使用该算法训练的检测器对MIT＋CMU 人脸测试库进行了测试，结果表明 该方法在训练速度、测试精度、检测时间等方面都优于相应的方法。

A fast face detection algorithm using enhanced AdaBoost based on Walsh features is proposed in this paper, and its training process is fast and works well under fewer non-face training samples.Firstly,the utility of Walsh features, instead of Harr-Like features can reduce the redundancy among features largely. Then, an enhanced double threshold AdaBoost algorithm is developed, where double threshold makes training process faster ; and in the process of training cascaded detector, the next classifier can be guided by the former classifier,which enhances the performance of the cascaded detector ;moreover,the adjustment to the threshold of each classifier can separate the training result of face and on-face as far as possible. Finally, the trained detector is tested on MIT + CMU test set, and experimental results show that its training speed, precision and detection time exceeds the corresponding method.