Research On Robust Face Recognition In Gated Boltzmann Machine

Face images are easy to access, informational, comfortable and high reliability, face recognition has broad application prospects in the field of identity verification. In recent years, face recognition has achieved satisfactory results under controlled conditions. However, in uncontrolled environment, face recognition is seriously challenged by variations in illumination, pose, expression, age and occlusion. In Our paper, we research on the detail of Deep Learning models which widely used. While Boltzmann Machines have been successful at unsupervised learning and density modeling of images, they can be very sensitive to noise in the data. Our work as follows:(1)We look into the structure and energy functions of Boltzmann Machine Models include:Restricted Boltzmann Machines, Deep Belief Networks, Convolutional Restricted Boltzmann Machines, Gassuian Restricted Boltzmann Machine, Gated Boltzmann Machines.(2)Analysis basic principles of the convolution neural networks (CNNs), introduce the training process and model structure of it. The convolution layer can make the original signal enhancement, and reduce noise as well as improve signal-to-noise ratio by convolution operation. It inspire us the convolutional layer added to the model is important and helpful.(3) As we want to solve the problem of Face Recognition variations in illumination, pose, expression and occlusion, we design the model called Convolutional Gated Boltzmann Machine.Our model is trained can learn the spatial structure of the occluders.Compared to standard algorithms, the Convolutional Gated Boltzmann Machine is significantly better at recognition and denoising on several face databases.