A Multi-biometric Face Recognition System Based on Multimodal Deep Learning Representations

A novel multi-biometric system for identifying a person’s identity using two discriminative deep learning approaches is proposed based on the combination of a convolutional neural network (CNN) and deep belief network to address the problem of unconstrained face recognition. CNN is one of the most powerful supervised deep neural networks, which is widely used to resolve many tasks in image processing, computer vision, and pattern recognition with high ability to automatically extract discriminative features from input images. This chapter proposes a multi-biometric face recognition approach based on multimodal deep learning termed fractal dimension transformation (FDT)-discriminative restricted Boltzmann machines (DRBM). It provides a brief description of the proposed approaches, including the FDT and DRBM approaches, which are used in the proposed local facial feature-based extraction approach. The most important hyper-parameter in the proposed FDT-DRBM approach is the number of the hidden units in the DRBM classifier.