基于加权最小平方误差boosting的人脸检测

A face Detection Based on a Weighted Least Squares Error Boosting Algorithm

近年来Adaboost算法被成功地用于人脸检测中,本文给出了一种基于加权最小平方误差boos-ting算法的人脸检测。首先本方法在每一次循环中用加权最小平方误差准则训练弱假设,与原始Ada-boost算法不同的是弱假设的生成不仅用于预测分类,而且用于估计每次预测的自信率,然后由这组含自信率的弱假设集成构造出强分类器。实践表明基于加权最小平方误差boosting算法的分类器有较高的检测率和较低的正样本误检率。

Recently AdaBoost algorithm was successfully applied to solve the problems of face detection. This paper presents a face detection based on a weighted least squares error boosting algorithm. First this method trains the weak hypothesis with the minimum weighted least squares error at each iteration. What differs from the original AdaBoost is that each weak hypothesis generates not only predicted classifications, but also estimate the confidence - rated of each of its predictions. Then, The final hypothesis is combination of the weak hypotheses whose predictions are confidence - rated. Experimental results prove that the classifiers based on a weighted least squares error boosting algorithm have a higher detection rate but a lower false positive rate.