A novel AdaBoost algorithm based on the multi exit asymmetric boosting and genetic algorithm is proposed in this paper. Both false acceptance rate and false rejection rate are fed back into current training stage, the cascade stages and the weight of local optimum weak classifiers are restricted by threshold comparison. The genetic algorithm with strong search ability is adopted for training local optimal weak classifiers, so most human faces in images and video sequences can be determined by less weak classifiers. Experimental results show that the proposed algorithm can achieve higher detection speed and more accuracy, effectively avoid overfitting and feature redundancy.%提出了一种基于多级联不对称增强和遗传算法的AdaBoost人脸检测方法。可把认假率和拒真率反馈给当前训练阶段，并通过阀值比较来控制级联层数和局部最佳弱分类器权值。用遗传算法训练选取的局部最佳弱分类器，实现用较少的弱分类器达到高检出率。仿真实验结果表明，该算法可以有效避免过拟合和特征冗余现象，获得较高的检测速度和精度。
展开▼
