Dropout Technique Based Convolutional Neural Networks Model for Face Recognition

From the past several years, face recognition is one of the challenging issues in computer vision, due to three main technical problems in it. 1) Expression problem: in which same person shows more than one expression. 2) Illumination problem: in which face images of the person are strongly corrupted by lightning and 3) poses problem: in which large portion of the face becomes invisible due to occlusion. To this end we propose a Deep Convolutional Neural Network architecture which consists of number of layers to overcome these challenging problems in face recognition. General discussion on deep learning is presented in the paper followed by brief introduction of various deep architectures. Detailed discussion on convolutional neural network is also given. Experimentation has been carried on the six benchmark face datasets varying in expression, illumination and poses and the results have been analyzed. The six face datasets on which we performed the experiments are ORL, Faces94, Yale, Yale-B, FERET, and CVL Faces. Performance comparison of the proposed approach with the already existing architectures i.e. Principal Component Analysis (PCA), Linear Discriminant Analysis (LDA), Discrete Cosine Transform (DCT), Independent Component Analysis (ICA) and Gabor Wavelet Transform (GWT) has been done.