Testing the impact of cloaked images on face recognition results
Cognitec has conducted repeated testing of software designed to obscure facial images through modifications that are scarcely perceptible to human viewers but intended to confound deep learning-based facial recognition systems. Two test results are presented demonstrating that both cloaking methods produced minor reductions from an optimal match score of 1.0, though they did not substantially alter the likelihood that the images represented the same individual.