一种改进型Adaboost算法的人脸检测

An improved Adaboost algorithm for face detection

原有的Adaboost算法在复杂背景及光源下,很容易出现人脸的误检问题,从而使人脸误检率较高。人脸相似区域的样本难以分类,导致出现权重过拟合现象使检测率降低。针对这些问题,本文提出了一种YCbCr肤色区域分割+改进型Adaboost算法的人脸检测算法。采用肤色区域分割排除复杂背景及光源的影响,将权重更新与正负样本误检率相结合,抑制人脸相似区域的权重过拟合现象,同时引入符合人脸的Haar-Like特征进一步提高检测率。通过实验证明,本文提出的算法在人脸检测中提高了检测率,降低了误检率和检测所需时间。

The original Adaboost algorithm is prone to face false detection under complex background and light source,which leads to a high rate of face false detection.It is difficult to classify the samples of face similar regions,which leads to the phenomenon of weight overfitting and reduces the detection rate.To solve these problems,this paper proposes a face detection algorithm based on YCbCr skin color region segmentation+improved Adaboost algorithm.Skin color region segmentation is adopted to exclude the influence of complex background and light source,weight updating is combined with positive and negative sample false detection rate,which could suppress the weight overfitting phenomenon in face similar areas.Meanwhile,Haar-Like features conforming to face are introduced to further improve the detection rate.Experimental results show that the algorithm proposed in this paper improves the detection rate,reduces the false detection rate and the detection time.