الفهرس | Only 14 pages are availabe for public view |
Abstract The emergence of screened films has led to an increased need for video content classification. However, a significant source of violence in these films can have a negative psychological impact on teenagers. Also, It’s important to filter sensitive content such as pornography and violence due to the increasing consumption of films by people of all ages. The use of deep learning in computer vision has shown great success and is currently receiving a lot of attention from researchers in this field. Inappropriate video content filtering has become a significant problem in modern society, particularly as internet access becomes more widely available. With the advancements in machine learning and neural network technology, many researchers have focused on creating models that can filter out pornography and other inappropriate content in movies. However, these models may not be as effective when it comes to filtering out inappropriate content in cartoons directed at children, as the filtering criteria for these types of videos are different from that for adult content. In this thesis, we propose a new CNN model called InspectorNet. InspectorNet is a deep neural network model designed to effectively detect and filter out inappropriate content such as pornography and violence in videos. This model utilizes both convolutional neural networks and transformer neural networks to overcome the limitations of current artificial neural network models. This research compares the performance of InspectorNet against ResNet and other previous methods used in this field, using a well-known animated cartoon images dataset called Danbooru2018 with a different number of classes. The result shows that InspectorNet outperforms ResNet in terms of classification accuracy. The comparison highlights that while InspectorNet requires significant computational resources for training, It shows better classification performance in inappropriate content filtering. |