Face Attribute Classifiers Based On Deep Neural Network

Face recognition is an important interdisciplinary subject which includes pattern recognition, computer vision, image processing and other fields. Face recognition has been widely used in real life because of its objective uniqueness. The study of multifunctional face recognition system can promote the development of face recognition in a certain degree. The key to improving face recognition performance is to collect massive labeled face dataset, construct discriminative face feature, and learn robust classifier.Deep learning is one of the most important branches of machine learning, which attempts to extract high-level features from multiple layers such as convolutional layers and pooling layers. So far there have been kinds of deep neural networks:convolutional neural network, deep belief network, recurrent neural network, which have been applied to computer vision, speech recognition, natural language processing and other fields. With the outstanding performance of Deep learning, more and more researchers have begun to use the deep neural network (especially the deep convolution neural network) to deal with face recognition and face detection.Based on the deep convolutional neural network, we propose a novel multi-task convolutional neural network (MTCNN) to build robust attribute representations.In order to reduce overfitting, we implement data augmentation to increase dataset manually. For model design, we put forward two ideas. Firstly, we treat attribute prediction as a regression task and employ Euclidean loss to optimize our models.Secondly,attribute prediction is taken as binary classification and we apply stochastic gradient descent (SGD) to minimize Cross Entropy Loss. We do extensive experiments to evaluate accuracy of 8 facial attributes. In average, attribute accuracy of our method can reach 96%.