Face age classification based on hybrid deep convolution neural network

To improve the accuracy of face age classification and reduce the time spent in age classification, a hybrid deep model based on fine-tuned deep convolution neural network (FDCNN) and probabilistic collaborative Representation classifier (PCRC) was proposed. First, on the IMDB dataset, the VGG-Face model was fine-tuned to a new model called the FDCNN;then the processed images were fed into the model for feature extraction and the PCRC was used to classify the output feature;finally, the performance of the deep hybrid model composed of FDCNN and PCRC was verified on the FG-NET, MORPH and CACD dataset. The verification shows that the average classification accuracy of PCRC is 4.6% higher than that of SVM, and the local feature from the activations of the penultimate layer can increase the accuracy of the classification compared with the activations of the last layer. The average classification accuracy of PCRC is 4.7% higher than that of SVM, and the mean absolute error (MAE) is 1.24, 0.14 and 0.06 lower than that of CA-SVR, DeepRank and DeepRank+ respectively. The average classification accuracy of the deep hybrid model is 3.6% higher than that of the classification model consisting of DCNN and SVM. Moreover, the comparison with the time distribution of each layer of VGG-Face model show that it takes less time for the deep hybrid model to operate. Therefore, the results show that the hybrid deep convolution neural network has good performance. © 2019, Editorial Board of Journal of Huazhong University of Science and Technology. All right reserved.