基于肤色特征和深度模型的人脸检测方法

A Face Detection Method Based on Skin Color Features and Deep Model

针对传统人脸检测方法中人脸特征提取过程复杂以及人脸检测准确性和鲁棒性不足的问题,提出一种将肤色模型与深度学习方法训练的深度网络模型结合进行人脸检测的方法。首先将图像转换到YCb Cr颜色空间,在该颜色空间下建立肤色的混合高斯模型,根据模型输出结果初步筛选出人脸肤色区域,并采用深度学习方法对筛选出的人脸肤色区域进行分类,该深度学习模型可以检测出不同姿态、表情、光线强弱下的人脸,再根据分类结果的置信度值进行判断,对部分置信度值满足条件的的分类结果再进行回归处理,从而获得更准确的检测定位。实验证明,该方法能明显提高检测精度和鲁棒性,并在一定程度上提升了检测速度。

This paper proposes a method that skin model combine with a deep network model trained by the deep learning method.The image is transformed into YCb Cr color space,and the mixed color Gauss model is set up in the color space,screen out skin color region preliminary.The screening result is input to the face detection model trained by the deep learning method to classify,according to the confidence value of the classification results,a part of the classification results with confidence value meet the conditions make the regression process,so as to obtain more accurate detection localization.