Children Inappropriate Content Detection Using Mask R- Cnn

Authors

  • Nithesh M, Naveen R, Naveen V

Abstract

The amount of inappropriate content on the Internet grows daily. Much of such content is unconstrained and freely-available for all users, requiring parents to adapt to parental control strategies for protecting their children. This is an attempt to mask or hide the data which are not applicable for the children. There are two major classes of data to be masked which have completely different features from one another (i) Adult content can be efficiently flagged as offensive based on the tradition and interests of the community who views it. The social-media applications such as tiktok and instagram allow users to post such content which is neither tagged as offensive content nor fall under age restriction which acts directly or indirectly as an threat to the adolescent aged community.Based on the above scenario it is evident and clear that we are compromising our culture and tradition and are moving towards the western culture. Certain content can be most appropriate to the western people and culture however it is not apt for the Indian culture.It could be more precisely said that the filter criteria algorithms do not understand the highly prideful tradition of the Indian and hence don’t filter efficiently based on our ideologies and practices.(ii) horror content Some movies like "SAW" contain some violent scenes which display blood and organs to the audience which are not appropriate for the children because on a high level such movies inculcates violent behaviour on the children and it could make the audience faint on sight of blood due to a medical condition called Vasovagal syncope. This as a combination rises and stands as a new and unsolved problem statement.This project detects the space in a frame which has inappropriate content for children and masks it by using a M R-CNN model.

Keywords: Image Processing, Mask R-CNN, Convolutional Neural Networks, Deep Learning, Machine Learning.

Downloads

Published

2020-05-18

Issue

Section

Articles