期刊名称：IAENG International Journal of Computer Science
印刷版ISSN：1819-656X
电子版ISSN：1819-9224
出版年度：2021
卷号：48
期号：2
语种：English
出版社：IAENG - International Association of Engineers
摘要：A face recognition algorithm based on VGG convolutional neural network with an improved pooling method is proposed. The recognition effect is not good as the recognized images may be interfered by various factors. For the situation that images are affected by illumination, the number of images are few and the quality of images are not good, based on the image pretreatment with normalization and de-average, the histogram homogenization is adopted to reduce the illumination effect, the randomly cutting out the number of images is to reduce the possibility of the network’s over-fitted and the Gabor wavelet transform is adopted to enhance the images. Then the Faster R-CNN network is adopted to carry out the face detection experiments on LFW database. Aiming at the problem that there are three full-connection layers in traditional VGG-16 network, which has a lot of parameters be produced in network training, the traditional VGG-16 network is improved by reducing the number of fully connected layers, replacing the original max-pooling method with a random square pooling method, which changes the last pooling layer to global mean pooling by referring to the GoogLeNet network method. Finally the simulation experiments are carried out on the LFW database and the self-built database. It is eventually found that the improved method effectively reduces the network training parameters, greatly reduces the network training time and obtains a good recognition rate.
