Spontaneous facial expression in unscripted social interactions can be measured automatically

摘要: Methods to assess individual facial actions have potential shed light on important behavioral phenomena ranging from emotion and social interaction psychological disorders health. However, manual coding of such is labor intensive requires extensive training. To date, establishing reliable automated unscripted has been a daunting challenge impeding development theories applications requiring expression assessment. It therefore essential that systems be developed with enough precision robustness ease the burden in challenging data involving variation participant gender, ethnicity, head pose, speech, occlusion. We report major advance spontaneous during an three strangers. For each (n = 80, 47 % women, 15 Nonwhite), 25 action units (AUs) were manually coded video using Facial Action Coding System. Twelve AUs occurred more than 3 time processed FACS coding. Automated showed very strong reliability for proportion AU (mean intraclass correlation 0.89), stringent criterion frame-by-frame was moderate Matthew’s 0.61). With few exceptions, differences detection related average pixel intensity small. Fewer 6 frames could but not automatically. These findings suggest progressed sufficiently applied observational research areas study.