A Compressed Deep Convolutional Neural Networks for Face Recognition

Recent years, deep convolutional neural network has led to significant improvements in face recognition and becomes one of the most popular techniques in computer vision community. However, deep CNN model requires vast amounts of data and time for training and deploying. To solve this problem, we present a deep compact convolutional neural network for face representation. First we apply PCA for initializing convolution filter. Then we adopt DCT and binary hashing for extracting face features. The models are trained on the CASIA-webFace datasets under Caffe framework. Experimental results show that the proposed methods achieve competitive accuracy on the LFW verification benchmark.