2020/01/16 by Vansh Narula, Narula, Vansh, Zhangyang +3
Computer Science · Psychology · #Computer Vision and Pattern Recognition (cs.CV) #Emotion and Mood Recognition #FOS: Computer and information sciences #Face and Expression Recognition #Face recognition and analysis #Machine Learning (cs.LG) #Machine Learning (stat.ML)
paper · pdf · doi:10.48550/arxiv.2001.06103
openalex publication_date 2020/01/16 · openalex created_date 2022/07/26 · openalex updated_date 2026/07/28
Image and video-capturing technologies have permeated our every-day life.\nSuch technologies can continuously monitor individuals' expressions in\nreal-life settings, affording us new insights into their emotional states and\ntransitions, thus paving the way to novel well-being and healthcare\napplications. Yet, due to the strong privacy concerns, the use of such\ntechnologies is met with strong skepticism, since current face-based emotion\nrecognition systems relying on deep learning techniques tend to preserve\nsubstantial information related to the identity of the user, apart from the\nemotion-specific information. This paper proposes an adversarial learning\nframework which relies on a convolutional neural network (CNN) architecture\ntrained through an iterative procedure for minimizing identity-specific\ninformation and maximizing emotion-dependent information. The proposed approach\nis evaluated through emotion classification and face identification metrics,\nand is compared against two CNNs, one trained solely for emotion recognition\nand the other trained solely for face identification. Experiments are performed\nusing the Yale Face Dataset and Japanese Female Facial Expression Database.\nResults indicate that the proposed approach can learn a convolutional\ntransformation for preserving emotion recognition accuracy and degrading face\nidentity recognition, providing a foundation toward privacy-aware emotion\nrecognition technologies.\n