Feature extraction is a critical technology that affects the accuracy of face recognition. However, certain features are highly related to changes in face are difficult to extract because of the influences of individual differences and illumination. Therefore, features can accurately describe the changes in face are urgently required. For this reason, this article proposes a feature extraction method based on deep learning. This method combines the features extracted by Local Binary Patterns and by Convolutional Neural Network convolutional layer in the network connection layer, thus obtaining classification features with high representation ability and solving the problem of single feature extraction. The VGG‐16 network proposed in this article has been improved by changing the framework structure. Some experiments based on the Labeled Faces in the Wild dataset are performed, and results show that, in terms of accuracy and the sensitivity to light, the proposed method reaches 99.56% and 80.35% respectively. The recognition results obtained from fused features are superior to which of single feature recognition. Simulation results show that the method is more robust to changes in the illumination condition and more efficient than the existing methods.

中文翻译：

---

基于深度卷积神经网络的人脸识别特征提取

特征提取是影响人脸识别准确性的关键技术。然而，由于个体差异和光照的影响，某些与人脸变化高度相关的特征难以提取。因此，迫切需要能够准确描述面部变化的特征。为此，本文提出了一种基于深度学习的特征提取方法。该方法在网络连接层结合Local Binary Patterns和Convolutional Neural Network卷积层提取的特征，从而得到具有高表示能力的分类特征，解决了特征提取单一的问题。本文提出的VGG-16网络通过改变框架结构进行了改进。基于Wild数据集中的Labeled Faces进行了一些实验，结果表明，在准确率和对光的敏感度方面，所提出的方法分别达到了99.56%和80.35%。融合特征得到的识别结果优于单特征识别。仿真结果表明，与现有方法相比，该方法对光照条件的变化具有更强的鲁棒性和效率。