基于深度学习的多任务人脸检测的设计与实现

Design and Implementation of Multi-task Face Detection Based on Deep Learning

笔者利用深度学习方法,基于AI框架Tensorflow和Mxnet,设计了3个级联网络,层层提取人脸特征,实现了多任务人脸检测。3个卷积神经网络级联完成了识别人脸与非人脸、获取人脸框坐标、获取人脸关键点坐标的3个任务。本研究成果一方面可用于图像或视频中的人脸检测,另一方面可对人脸进行对齐矫正,为后续人脸身份识别与人脸属性识别的研究提供数据支持。

The authors use deep learning methods,based on the Tensorflow and Mxnet which is AI frameworks,design three cascaded networks to extract facial features layer by layer,and it implement multi-task face detection.The three cascaded convolutional neural network work on the three tasks which is identifying faces and non-faces,obtaining face frame coordinates,and obtaining key point coordinates of the face at the same time.On the one hand,the research results can be used for face detection in images or videos,on the other hand,it also works well on face alignment or face correction,it could also provide data for subsequent research of face identification and face attribute recognition.