摘要
为了提高复杂光照条件下的人脸检测识别率,提出了一种基于Retinex图像增强技术应用于多任务卷积神经网络(multi-task cascaded convolutional networks,MTCNN)的人脸测算法。算法用Retinex理论对图像进行增强,能明显提高MTCNN在不同光照场景下的人脸检测精度,同时使面部五个关键点的定位更准确。实验证明,在复杂光照场景下,该方法比原始MTCNN网络的人脸检测具有更好的效果,有利于后期的人脸对齐及分类任务。
In order to improve the recognition rate of face detection under complex lighting conditions,a face detection algorithm based on Retinex image enhancement technology applied to multi-task cascaded convolutional networks(MTCNN)was proposed.The algorithm uses Retinex theory to enhance the image,which can significantly improve the face detection accuracy of MTCNN in different lighting scenarios,and make the positioning of the five key points of the face more accurate at the same time.Experiments have proved that this method has better results than the original MTCNN network in face detection in complex lighting scenes,and is beneficial to the later face alignment and classification tasks.
作者
薛晨
宁志刚
XUE Chen;NING Zhigang(School of Electrical Engineering, University of South China, Hengyang, Hunan 421001, China)
出处
《南华大学学报（自然科学版）》
2021年第3期70-74,共5页
Journal of University of South China：Science and Technology
