Automatic Face Recognition

The tremendous growth in the number of surveillance cameras as well as cameras in mobile devices (such as smartphones and notebooks) has generated a huge interest in automatic face recognition. An important topic in face recognition is heterogeneous face recognition (HFR), where the two faces to be matched have been captured from different sensing modalities. Matching a sketch of a suspect drawn by a forensic artist to a large collection of mug shot photos is one example of HFR.

University Distinguished Professor Anil Jain and Ph.D. candidate Brendan Klare have utilized the ICER resources for their research on HFR. Since face images are generally represented by feature descriptor vectors with a dimensionality (number of features) in the order of thousands, the computing resources provided by ICER have been instrumental in both the statistical learning and matching stages of their experiments. Through ICER resources, Klare and Jain have been able to leverage a mug shot database of more than 1 million face images. This large data set has also allowed them to investigate the effects of age, gender, and ethnicity on matching performance.