Improving signal detection in software-based facial expression analysis

Algorithmic and equipment-based procedures for emotion detection are often afflicted by measurement error or signal noise. In this paper, we analyze the signal-noiserelation of software for automated facial expression analysis used to measure emotional response to marketing stimuli. We isolate the noise and discuss, apply, and evaluate several methods for reducing the noise. The results show that noise is a challenge in automated analysis of facial movement data, but can be reduced by applying fairly simple methods. Using data from a real market research study we show that noise can be reduced to a negligible level.