Exploiting Sound Latency Using Low-level Affective Video Features in Amateur Video

Abstract

The increased availability of multimedia equipment has resulted in large repositories of publically available, amateur videos. Users need to be able to identify and retrieve videos that contain content of interest. Automated methods are desirable, as manual content discovery is tedious. One of the more difficult challenges in video indexing is affective video indexing, which focuses on content that is intended to evoke certain emotions in users. Without pre-determined cinematographic cues, indexing strategies applied to amateur videos must rely on the exploration and analysis of low-level characteristics such as sound energy and motion. This study focuses on (1) improving the existing method for correctly identifying target affective content, and (2) exploiting sound latency in correlation to object motion in amateur videos containing slapstick, one of the most popular humor techniques. By utilizing low-level video characteristics, the identification of the targeted content can be performed without relying on the emotional responses of humans.