Disguised Face Recognition

Problem Definition

Recognizing a person’s face images with intentional/unintentional disguising effects such as make-up, plastic surgery, artificial wearables (hats, eye-glasses) is a challenging task, termed as Disguised Face Recognition (DFW). Recently, this task has gained attention because of the pandemic situation where people are wearing masks.

Approach

In this project, We propose a Feature EnsemBle Network (FEBNet) for recognizing Disguised Faces in the Wild (DFW). FEBNet encompasses multiple base networks (SE-ResNet50, Inception-ResNet-V1) pretrained on large-scale face recognition datasets (MS-Celeb-1M, VGGFace2) and fine-tuned on DFW training dataset.

Acknowledgements

This project was done as part of my internship at Computer Vision Lab, IIT Madras under the guidance of Dr Anurag Mittal. You can find the research paper explaining the methods and experiments here.