PROJECT TITLE :

Mutual Component Convolutional Neural Networks for Heterogeneous Face Recognition

ABSTRACT:

For example, the goal of heterogeneous face recognition (HFR) is to recognise people in photos that are both visible and near-infrared. There is a huge disparity in the number and type of training samples available for HFR. The MC-CNN, a modal-invariant Deep Learning framework, is proposed in this research to address both of these concerns simultaneously. As an FC layer, mutual component analysis (MCA) is incorporated into modern deep CNNs by our MC-CNN, which is a generative module. This FC layer is meant to extract modal-independent hidden components and is updated based on the maximum likelihood analytic formulation instead of back propagation, which naturally minimises overfitting from limited data.. The network for modal-invariant feature learning is also updated using an MCA loss. Our MC-CNN outperforms numerous fine-tuned baseline models significantly, according to extensive experiments. On the CASIA NIR-VIS 2.0, CUHK NIR-VIS, and IIIT-D Sketch datasets, our approaches outperform the state-of-the-art.