Signal Processing and Communications Laboratory

Department of Engineering

Disguised Face Identification Research

Amarjot has developed the next generation face recognition system that is able to identify faces covered with disguises such as scarfs, beards, caps, and glasses. Individuals on the criminal databases around the world often disguise their faces with hats and beards to avoid identification by face recognition systems. This system will help to tackle the global security challenges by identifying masked criminals that pose serious real-world threats, in turn, making society more secure.

Citation: A Singh, D Patil, GM Reddy, SN Omkar, Disguised Face Identification (DFI) with Facial KeyPoints using Spatial Fusion Convolutional Network, IEEE International Conference on Computer Vision Workshop (ICCV), 2017

Algorithm

The proposed AI model uses a convolutional neural network to remember how the structure of the face should look like from the labeled training dataset. Using this the AI model locates for points on the disguised that define the facial structure. The input to the DFI model are disguised images with coordinates of the 14 keypoints, marked manually by humans on the disguised face. The Deep Convolutional network looks at a batch of these disguised images and learns to predict the keypoints for the parts of the face (hidden or otherwise). The network tries to correct the prediction error made for each keypoint and eventually is able to predict the keypoints reasobaly well on the seen images. The angle between the key points located on the face is finally used to Identify the disguised face.

The images below shows the points to be located on the face and an example image on which our proposed algorithm locates the required points:

Fig-1n.jpg im.jpg

Media Coverage

Cover of The Economist

DFI-Cover.jpg

wb.jpg