In Partial Fulfillment of the Requirements for the Degree of Master of Science
will defend her thesis
Privacy Preserving Face Matching Using Frequency Components
A deep look into research today projects the efforts of scientists and researchers to mimic a human itself. The goal of mimicking a human involves devising an automated system that can be as perfect as the human. The research along these goals started as early as the 1980s leading to the birth of computer vision. Object detectors play a fundamental role in computer vision by trying to perfect the answer to the very question-- “where is the object located?”. The computing and cost limitations have been lifted by an increase in computational power and a decrease in the cost of the performance, leading to state-of-the-art object and face detectors. The availability of such resources has led to a vast increase in the use of object and face detectors in real-time applications such as autonomous driving, video surveillance, robot vision and many more. With an increase in the popularity of object and face detectors, there has been an increase in the privacy issues related to the same. Faces construct a very sensitive portion of an image, which in this age has become vulnerable with increasing use of the applications involving object detectors. Increasing concerns over the privacy of people have led to the research involving anonymizing faces, privacy-preserving feature selection, and active learning. This thesis aims to provide a possible solution to the privacy issues by utilizing the frequency components of an image, instead of using the intensity components which are relatively easier to reconstruct and comprehend by an attacker. The process involves calculating scores for the metrics-- Cosine, Correlation, Manhattan-- between two image subjects. The score is tested against the threshold values corresponding to each metric to make a global decision of whether the image subjects are the same or not. We also analyze the role DCT coefficients play in making the combined decision.
Date: Friday, November 22, 2019
Time: 10:00 - 11:00 AM
Place: PGH 550
Advisor: Dr. Shishir K. Shah
Faculty, students, and the general public are invited.