一种复杂背景下的人脸检测方法

摘要

Aiming at resolving the problems of face detection algorithm under complicated background such as low detection rate and high false-alarm-rate etc, an effective method of face detection in color images based on improved AdaBoost algorithm and skin color is presented. By improving parameters of sample weights and weak classifier weighting value, the traditional AdaBoost algorithm has been improved. It effectively restrains the excessive increase of weights of hard samples,strengthens the capacity of classifier for recognition of samples and the system detection rate is increased. The detection result is screened with YCbCr space skin color model to filter error-detected “non-face” area. The experimental results indicate that this method assures the face detection speed and precision of improved AdaBoost algorithm under complicated background, decreases the false-alarm-rate and has strong robustness.%针对在复杂背景下现有人脸检测算法存在检测率低和误检率高等问题,提出了一种基于改进AdaBoost算法和肤色校验相结合的彩色图像人脸检测方法.首先对传统AdaBoost算法进行了改进,通过改进样本权值参数和弱分类器加权参数,有效地抑制了困难样本权值的过分增大,加强了分类器对样本的识别能力,并提高了系统的检测率;然后将AdaBoost算法检测结果用YCbCr空间的肤色模型进行筛选,过滤误检的非人脸区域,进一步降低误检率.实验结果表明:该方法在复杂背景下极大地保证改进AdaBoost算法的人脸检测速度和准确性的同时,降低了误检率,具有较好的鲁棒性.