Disguised Face Recognition using Convolutional Neural Network

In our day-to-day life, we are implementing various biometric technologies especially facial recognition to ensure privacy of our information as well as for secured access. But facial recognition has become a tough task recently as faces could be covered with various disguises. In this paper, disguised face recognition has been taken into consideration by utilizing Convolutional Neural Network (CNN) approach. CNN architectures have a lot of variations optimized for various purposes. FaceNet is one of them and performed really well on facial image dataset. The main purpose of this work is to recognize the human faces covered with various disguises like sunglass, hat, beard, mustache, hair, scarf etc. Multi Task Cascaded Convolutional Network (MTCNN) has been taken into account to detect facial regions and FaceNet architecture has been utilized to find facial embeddings. After that, Support Vector Machine (SVM) has been used to categorize the similarity of faces. Our model achieved an accuracy of 88.78% on IIIT-Delhi Disguise Version 1 Face Dataset and the result has been compared with the state-of-the-art architecture. It has been shown that our model surpassed the state-of-the-art architecture.