复杂背景中人脸检测与眼睛定位

Human Face Detection and Eyes Location in a Complex Background

本文对Viola等人提出的基于类Haar特征及AdaBoost的人脸检测算法进行了改进,将原来的单一阈值弱分类器改为输出几乎可模拟任意概率分布的特征查找表(FST)弱分类器,构建出人脸检测模块;同时,训练左、右人眼分类器对输入图像进行眼睛定位,并且利用ROC曲线对改进算法收敛速度、人脸检测器和人眼检测器的性能与Viola-Jones算法进行了比较分析。结果表明,改进后的算法具有较高的检测率和较低的误警率。

The face detection method presented by Viola et al based on the Haar-like feature and the AdaBoost algorithm is improved. The original weak classifier with a single threshold value is substituted by Feature Search Table （FST） whose outputs can be suitable for almost any distribution. And we use the left and right eye classifier simultaneously for eyes locating. And we demonstrate the value of the ROC curves in evaluating and comparing the modified algorithm with the Viola-Jones algorithm in convergence rate,the capability of the face detector and the performance of the eye detector. It shows that the modified algorithm is better than the Viola-Jones algorithm, and it helps us improve the hit rate, and makes less falsc-alarrns rate.