一种改进的AdaBoost人脸检测算法

为了进一步提高AdaBoost算法的检测准确率和解决AdaBoost算法的退化问题,提出了一种基于线性不对称分类器(LAC)的改进AdaBoost人脸检测算法。该算法对样本权重的更新规则进行了调整,当训练节点分类器的前m个弱分类器时,由于样本权重更新合理,采用原始权重更新方法;当已训练产生一定数量的弱分类器序列后,退化问题严重,由LAC算法对前期训练获得的弱分类器序列进行学习形成最优强分类器,计算该强分类器对负样本集的错分率FPR,更新样本权重,依次循环直到完成该节点所有弱分类器的训练并得到最佳节点分类器,最后通过级联各个节点分类器,构成人脸检测分类器。同时,对已有的Haar特征进行了改进和完善。实验结果表明,该方法不仅提高了检测精度,而且抑制了退化现象,使人脸检测分类器的性能得到进一步的提高,具有较大的实用价值。

An improved AdaBoost face detection algorithm

To further improve the detection accuracy of AdaBoost algorithm and solve the degraded issues, an improved Adaboost face detection algorithm based on LAC is presented in this paper. The proposed algorithm updated sample weight by two steps. First, training the first weak classifiers of node classifier by traditional weight updating method; Second, when a certain number of weak classifiers sequence is generated, degradation problems become serious, then the weak classifier sequences which have already been acquired are applied to form strong classifier by LAC algorithm, and the (false positive rate)of Negative samples were calculated by strong classifier, the sample weight will be affected by FPR. Repeat these steps until the training of all the node weak classifiers is completed, then the best node classifier is got. At last, by cascading all the best node classifiers, the face detector is acquired. Meanwhile, the existing Haar features are improved and perfected in this paper. The experiment results show that the method not only improves the detection precision, but also inhibits the degradation phenomenon, which greatly improve the face detection performance.