Low level visio-temporal features for violence detection in cartoon videos

Abstract

Cartoons are an informative way for creating awareness; children take keen interest in watching cartoons and spend leisure time in front of television. Unfortunately there is an inclination towards violence and other objectionable scenes in cartoon videos that have very bad impact on the developing personality of children. Extensive use of such violent scenes is one of the factors of increase of violence in society. Parents rely on cartoons as a mean of amusement for their children without analyzing its content which can pose serious negative impacts on child behavior. To protect children from violent content means are require for automatic detection of violence content in cartoon videos. In this paper different low-level visual features are evaluated for violence detection in cartoon videos using indigenously developed dataset that is categorized as violent and non-violent. From our results it has been observed that low-level visual features could not be used efficiently for the detection of violence. However these features are very helpful in identifying the character and situation that if combined with a knowledgebase capability can be used for detection of violence and other objectionable elements in the cartoon videos.