Local Receptive Field-Extreme Learning Machine based Adult Content Detection

Adult contents have become available everywhere whether in social networks, TV channels and websites. Children protection from pornographic contents is required in all societies and environments. Inappropriate visual contents have an impact on children's psychological development. Parents' censorship is important to solve the problem but this task is time consuming and needs full time employment for activities monitoring. Automatic censorship is highly required to be integrated in all types of media such as TV channels. This paper proposes an automated censorship system that is able to be implemented on low cost embedded system. The public dataset NPDI was used to train and evaluate the performance of the proposed deep model for visual contents detection. The key frames selected from videos are used for porn/non-porn classification. Efficient features have to be extracted in order to perform accurate classification. The paper utilizes a local Receptive Field-Extreme Learning Machine (LRF-ELM) model which is a fast deep learning model to extract features and map them to specific classes in a short time. Our proposed method was found to give good performance in terms of accuracy and training time. The accuracy of 82.9% was achieved by using a normal Central Processing Unit (CPU).