全卷积神经网络的多尺度人脸检测的研究

为实现快速而准确的人脸检测，提出了一种基于全卷积神经网络的多尺度人脸检测的方法，将卷积神经网络模型AlexNet的全连接层改为全卷积层，并将分类层改为人脸与非人脸的二分类，训练之后准确率达到99.16%。将训练好的分类模型用于人脸检测时，待检测图片通过多尺度变换后输入全卷积网络得到特征图的概率矩阵，用非极大值抑制得到最精准的人脸框。检测结果表明，该方法在人脸检测时准确率高，检测时间短，表现出较好的性能。

Multi-Scale Face Detection of Full Convolution Neural Network

In order to achieve fast and accurate face detection, a multi-scale face detection method based on full Convolutional Neural Network（CNN） is proposed. The full connectivity layer of the convolutional neural network model AlexNet is changed to full convolution layer, and divided the layer into two categories of face and non-face, the accuracy after training as high as 99.16%.When the trained classification model is used for face detection, the image to be detected is input to the full convolutional network through multi-scale transformation to obtain the probability characteristic figure, and the most accurate face frame is obtained by the inhibition of non-maximal value. The test results show that this method has the advantages of high accuracy, short detection time and good performance in face detection.