Researchers recognized for face image generation work
A team of ECE researchers received the Best Paper Award at the 2018 IEEE International Conference on Identity, Security, and Behavior Analysis (ISBA), held in Singapore. The paper, "Normalized Face Image Generation with Perceptron Generative Adversarial Networks" introduces a novel adversarial network to generate the frontal and neutral face image and proposed a "recognition via generation" approach to automatically generate the hard negative samples of the inputs to improve the facial expression recognition performance. The paper was authored by Xiafeng Liu, a visiting Ph.D. student in electrical and computer engineering, and co-authored by Vijayakumar Bhagavatula, U.A. and Helen Whitaker Professor of Electrical and Computer Engineering, Yubin Ge (CMU), Chao Yang (USC), Jane You (HK PolyU), and Ping Jia (UCAS).
“This is my first Best Paper Award, and could be a encouragement for my future works about the general face recognition related area,” says Liu. “As a visiting Ph.D. student at Carnegie Mellon, Professor Bhagavatula made a great effort to instruct the experiments and the oral presentation details.”
Abstract
This paper presents a deep neural architecture for synthesizing the frontal and neutral facial expression image of a subject given a query face image with arbitrary expression. This is achieved by introducing a combination of feature space perceptual loss, pixel-level loss, adversarial loss, symmetry loss, and identity-preserving loss. We leverage both the frontal and neutral face distributions and pretrained discriminative deep perceptron models to guide the identity-preserving inference of the normalized views from expressive profiles. Unlike previous generative methods that utilize their intermediate features for the recognition tasks, the resulting expression- and pose- disentangled face image has potential for several downstream applications, such as facial expression or face recognition, and attribute estimation. We show that it produces photorealistic and coherent results, which assist the deep metric learning-based facial expression recognition (FER) to achieve promising results on two well-known FER datasets.