(IJACSA) International Journal of Advanced Computer Science and Applications, Vol. 12, No. 5, 2021 A Novel Pornographic Visual Content Classifier based on Sensitive Object Detection Dinh-Duy Phan 1 , Thanh-Thien Nguyen 2 , Quang-Huy Nguyen 3 , Hoang-Loc Tran 4 , Khac-Ngoc-Khoi Nguyen 5 , Duc-Lung Vu *6 Faculty of Computer Engineering, University of Information Technology Vietnam National University Ho Chi Minh City, Viet Nam Abstract—With the increasing amount of pornography being uploaded on the Internet today, arises the need to detect and block such pornographic websites, especially in Eastern cultural countries. Studying pornographic images and videos, show that explicit sensitive objects are typically one of the main charac- teristics portraying the unique aspect of pornography content. This paper proposed a classification method on pornographic visual content, which involved detecting sensitive objects using object detection algorithms. Initially, an object detection model is used to identify sensitive objects on visual content. The detection results are then used as high-level features combined with two other high-level features including skin body and human presence information. These high-level features finally are fed into a fusion Support Vector Machine (SVM) model, thus draw the eventual decision. Based on 800 videos from the NDPI-800 dataset and the 50.000 manually collected images, the evaluation results show that our proposed approach achieved 94.06% and 94.88% in Accuracy respectively, which can be compared with the cutting-edge pornographic classification methods. In addition, a pornographic alerting and blocking extension is developed for Google Chrome to prove the proposed architecture’s effectiveness and capability. Working with 200 websites, the extension achieved an outstanding result, which is 99.50% Accuracy in classification. Keywords—Computer vision; image proccessing; object detec- tion; pornographic recognition and classification; blocking exten- sion; machine learning; deep learning; CNN I. I NTRODUCTION In the digital era, information has become a powerful weapon to manipulate the development of a society. People nowadays are easy to find and upload any information they want on the Internet. On the one hand, these pieces of informa- tion sometimes are good and bring positive value to the human race. On the other hand, many harmful kinds of negative information can also be found with only some keywords by anybody, even children. Pornography content is one of them. Many women worldwide are victims of sexual cybercrimes because their private videos are spread on the social network. Furthermore, pornographic content is even restricted in many countries. From the above problems, the need for an effective pornographic visual content detector is necessary. Many efforts have been made recently to classify porno- graphic images among normal ones. In the early stages, the skin-based method has been applied. These approaches check whether images have nude people or not based on the ratio of the exposed skin. Another approach is handcrafted features- based, which uses various descriptors to extract key point low-level features in an image. A visual codebook may be learned by applying the k-means algorithm on a training set. After that, the trained codebook could represent any images, and a classifier may detect pornographic ones. Applying low- level features to identify obscene images, it achieved signifi- cantly higher performance than skin-based methods. However, representing images by visual words still suffers a severe problem since it ignores the spatial relationship, which is very important to represent the image’s content. The state of the art approaches for this classifying problem are based on deep learning methods. These approaches build models with neural networks that let it learns features from the image’s contents itself. Previous studies [1], [2] [3], have implemented the above approaches and achieved some particular success, especially the deep-learning-based approach has proposed a potential development for this problem. However, the context in images is very complicated, and there are many similarities between a pornographic image and a normal one. The normal image that contains a large region of exposed skin (e.g., swimming, wrestling, people wearing bikinis) or contains people with sexy poses may be misclassified as pornography. Misclassification may seriously affect the user experience while using the internet. To avoid this problem, we have to clarify what is pornography. According to Oxford Advanced Learner Dictio- nary, “Pornography is magazines, DVDs, websites, etc., that describe or show naked people and sexual acts to make people feel sexually excited, especially in a way that many other people find offensive.” 1 From the definition, an image can be determined as pornographic if it contains naked people. In other words, pornographic images are images that consist of human’s sensitive objects and organs such as breasts, anus and genitals. We called this method is the sensitive object-based approach. Based on that insight, this paper presents a novel approach for pornographic content detection and classification, which not only leverages the advantages of previous approaches but also compensates for these methods’ weaknesses. Our main ap- proaching strategy is using the effectiveness of object detection to identify pornographic elements in visual content with steady prediction. Additionally, skin and human recognition are also integrated into our method to distinguish between images with humans from images without humans, but with human like skin colors such as sand or wood. These two modules not only capable to augment the classification’s decision but also can be served as the counterweight to prevent the potential bias that comes from object detection. Ultimately, a linear classification 1 https://www.oxfordlearnersdictionaries.com/definition/english/pornography www.ijacsa.thesai.org 787 | Page