Multi-level detector for pornographic content using CNN models

This paper focuses on detecting and classifying pornographic content (images and videos) by using a multi-level CNN model with some supportive models. The main approaching method is to determine the images (keyframes extracted from videos) containing sensitives content or not by applying object detection model Mask R-CNN, which is the completely new approaching method in pornographic recognition. Moreover, the proposed model also adapts some other methods such as feature extraction and classifying based on CNN to increase the accuracy of the adaptive methods and ignore non-pornographic images and videos. Experimental results using the Pornography-800 and Pornography-2K datasets, performance of our method is reaching the accuracy of 92.13% and 90.40% respectively, show the effectiveness of the proposed method.