An efficient Deep Learning of scarce and diverse faces for Face recognition

Face recognition (FR) is a challenging task when the samples per person are scarce and that too with each sample belonging to different facial expressions. A good classifier requires sufficient large numbers of training samples to train and produce accurate output when subjected to test samples. Nowadays, Deep learning (DL) is been extensively used where numerous handling layers tend to acquire records with numerous stages of feature at many levels. The use of Deep learning has radically upgraded the state of art presentation and solved many existent domain presentations. The main contribution of this work is the way 18 features corresponding to four frequencies, four orientations and two with spatial context for each individual pixel are extracted using Gabor filtering and finally reducing the 18 dimensional feature spaces to 1 dimensional feature vector using principal component analysis (PCA). This work uses a small YALE dataset of fifteen subjects with eleven facial expressions. Deep learning with Convolutional neural network (CNN) is used to train 154 samples in random and the remaining 10 samples were subjected for testing the CNN. The classification result obtained with randomly selected test sample was 100% for single test sample and 95% recognition rate when 10 samples are subjected to test.