基于Inception-v3卷积神经网络模型的人脸识别

Face recognition based on Inception-v3 convolutional neural network model

实现了一种基于Inception-v3卷积神经网络的人脸识别方法。该方法通过修改pool_3以上的结构来改变网络深度,从而使卷积神经网络模型能自动提取人脸特征并进行分类。利用OpenCV自建人脸数据库进行训练,通过批量梯度下降法不断优化参数,利用Dropout方法解决过拟合问题。结果表明:该方法在自建人脸数据库上的图像识别率达到98%,并且对光照强度、面部表情变化等干扰具有鲁棒性。

A face recognition method is implemented based on Inception-v3 convolutional neural network.The method changes the network depth by modifying the structure above pool_3,so that the convolutional neural network model can automatically extract facial features and classify them.It is proposed to use OpenCV self-built face database for train-ing,optimize parameters by batch gradient descent method,and use Dropout method to solve over-fitting problem.The experimental results show that the image recognition rate of the proposed method on the self-built face database can reach 98%,and it is robust to disturbances such as illumination intensity and facial expression changes.