Facial feature extraction method based on fast and effective convolutional neural network

It is difficult for shallow networks to extract effective facial features, while it is time-consuming using deep networks with many iterations. This paper proposes a facial feature extraction method based on Fast and Effective Convolutional Neural Network (FECNN). Firstly, a deep network based on NIN method is presented to extract effective facial features. Then, a new Inception structure is used to deepen and widen the network while reducing the number of parameters. Finally, this network is embedded with Batch Normalization (BN) algorithm which greatly accelerates network convergence. Experimental results indicate that FECNN converges efficiently and robust facial features are extracted with less parameters.