Gore Image Automatic Censorship Pro-gram

Abstract

A Matlab-based automatic gore image censoring program is presented in this research. The program uses Alexnet to classify set of images as gore images and non-gore images. This pre-trained convolutional neural network is further trained to allow more dynamic applications that is able to successfully classify gory images. A gore image is an image that contains violent scenes such as bloods from human bodies and mutilated human bodies that may be disturbing to some audiences especially children.  Four sets of images are used as such gore images, non-gore images, confusing non-gory images and a gore images with low threshold of blood. A total of 32773 images are tested. Results show that the modified Alexnet program can censor gore images and has an accuracy of at least 98.43% with a censoring rate of 20 seconds per image. Most of the errors in the classification process are due to low blood threshold and confusing non-gore images. This program can be used to improve the censorship capabilities of news media and other publication firms