2018/03/19 by Brandon RichardWebster, RichardWebster, Brandon, So Yon Kwon +7
Computer Science · Psychology · #Face recognition and analysis #Face and Expression Recognition #Emotion and Mood Recognition
paper · pdf · doi:10.48550/arxiv.1803.07140
Scientific fields that are interested in faces have developed their own sets\nof concepts and procedures for understanding how a target model system (be it a\nperson or algorithm) perceives a face under varying conditions. In computer\nvision, this has largely been in the form of dataset evaluation for recognition\ntasks where summary statistics are used to measure progress. While aggregate\nperformance has continued to improve, understanding individual causes of\nfailure has been difficult, as it is not always clear why a particular face\nfails to be recognized, or why an impostor is recognized by an algorithm.\nImportantly, other fields studying vision have addressed this via the use of\nvisual psychophysics: the controlled manipulation of stimuli and careful study\nof the responses they evoke in a model system. In this paper, we suggest that\nvisual psychophysics is a viable methodology for making face recognition\nalgorithms more explainable. A comprehensive set of procedures is developed for\nassessing face recognition algorithm behavior, which is then deployed over\nstate-of-the-art convolutional neural networks and more basic, yet still widely\nused, shallow and handcrafted feature-based approaches.\n