Research on Face Detection and Face Attribute Recognition based on Deep Learning

In recent years, facing recognition technology has been widely used in video surveillance, image retrieval, advertising, human-computer interaction and other fields. In terms of face detection, the traditional image classification model has the problems of long training time, huge amount of parameters and huge amount of calculation. To resolve these problems, this paper proposes a gender recognition and age estimation algorithm based on multi-task learning,with improving the classification model that achieves excellent performance in the image classification field. Using Convolutional Neural Networks can extract preliminary feature, and then join the multi-task learning layer, performs gender recognition and age estimation tasks by guiding the weight update of the network through the backpropagation algorithm. In this paper, by designing an efficient neural network structure for multi-attribute learning of faces, it realizes that multi-attributes share features in the shallow layer, and reduces network overfitting. It also learns features independently at high levels, and improves model accuracy. The algorithm can predict static face images and the dynamic video, and can also perform multi-target detection and multi-attribute recognition.