基于卷积神经网络特征融合的人脸识别算法

为提高卷积神经网络的识别性能,提出了一种基于多种卷积神经网络模型的特征融合方法。论文通过构建一个深度学习网络,将多种卷积神经网络模型如ResNet、InceptionV3和VGG19提取的特征进行融合,并将融合后的特征应用到人脸识别中,据此训练出特征融合网络模型的网络参数;最后利用计算求出的阈值来区分类别。实验结果表明,在人脸库LFW数据集上,论文算法的人脸识别率可达98%;与现有的单一卷积神经网络相比,论文算法识别率更高。

Face Recognition Algorithm Based on Convolution Neural Network Feature Fusion

In order to improve the recognition performance of convolutional neural network,a new feature fusion method is proposed based on multiple convolutional neural network model.Feature fusion is one of the important methods to improve the perfor mance of convolutional neural networks.In this paper,a deep learning network is constructed to fuse various CNN features,such as ResNet,InceptionV3 and VGG19.And the fused feature is applied to face recognition,thereby training network parameter of fea ture fusion network model.Finally,the calculated threshold is used to distinguish the categories.Experimental results show that the recognition rate of the feature fusion algorithm can reach 98%in the face database LFW data set.Compared with the existing single convolutional neural network,the recognition rate of the algorithm is higher.