Video digest based on heart rate

In video digesting, not only features and keywords extracted from a content itself but viewer's input are essential to incorporate subjective impressions and perceived importance. The present study aims at providing heart rate based arousal level as weights for determining shots that are deemed impressive and consuming. Heart rate is suitable for video content analysis, as it can be measured without being obtrusive and can be the source of different types of psychophysiological indices. Based on the findings that heart rate decelerates when emotionally arousing images are shown, we had investigated the fidelity of heart rate in detecting impressive shots. Three video contents with different valence/arousal levels were presented to the subjects, and the heart rate based "best shot" selections were compared against subjects-selected shots. The resulting hit ratio was 37% (the probability of random guessing is 17%). This result is about the same as other physiology-based affect recognition mechanisms with one class of input. We can expect this fidelity to be improved further by combining other heart rate based measures such as heart rate variability.