Detection of Violent Content in Cartoon Videos Using Multimedia Content Detection Techniques

Abstract

Children are the most vulnerable to ideas presented in Cartoon videos and TV. Cartoons have become one of the most important source of entertainment, but it also introduce a lot of ideas that are not suitable for them. Violence is one of the unwanted feature that is prevalent in cartoons to put element of fantasy and enchantment. In order to stop children from viewing violent intense cartoons, the best strategy is to make them inaccessible. Therefore, some sort of filters should be placed at certain hubs to perform this task. The challenge is that how a filter will know that a particular cartoon video has violent content in it. The meta-data telling the world about the video does not inform that the video consists of violent material. Certain frames/snapshots/images of video, if analyzed using image processing techniques, can help in concluding that a particular video has intense material in it. The aim of this work is to classify social media videos especially related to animated cartoons with violent / nonviolent behaviors. It addresses the problem of content based image matching algorithms based on key point descriptors. The basic goal is to extract general information from an image without any specific query. First SIFT-descriptors are extracted from a large set of images. This set of descriptors are then defined as a means of providing fast and accurate comparisons between images and distinguish between violent and nonviolent images in combination with Machine Learning algorithms. The results are then compared for each classifier with varying parameters.