Partial Occlusion Face Recognition Based on CNN and HOG Feature Fusion

Abstract

Although the present studies of face recognition have relatively been mature, in some complex scene environments, the efficiency of face recognition needs to be improved due to the influence of uncertain factors such as changes in illumination, changes in facial expressions, and partial facial occlusion. In order to improve the efficiency of face recognition, this paper proposes a feature fusion method based on convolutional neural networks (CNN) model and hog model. The model extracts rich implicit features from the original image by using convolutional neural network (CNN), and uses Dropout technology in the convolutional layer and the fully connected layer to randomly inhibit the activation of some neurons, so as to better solve the problem of overfitting. Moreover, this method also gives full play to the stability and robustness of Histogram of Oriented Gradients (HOG) Feature Enhancement Model. After extracting the CNN features and HOG features of the face, the method combines CNN SoftMax and HOG-SVM classifiers. The experimental results show that the recognition rate of this method is higher than that of single convolution neural network, which can reach 96.1%.