Application of Pornographic Images Recognition Based on Depth Learning

Abstract

With the rapid development of the Internet, the images become the main medium of information dissemination, while the spread of pornographic images are getting more serious. Therefore, we propose a detection method of pornographic images based on a combination of global and local features. Considering the NPDI database's defective both in quality and quantity, so this paper constructs new database CUC_NSFW (Not Suitable for Work) applying data augmentation methods to improve the classification performance. Pornographic images with only exposed sensitive organs become the bottleneck of improving model recall ratio. We design a sensitive organs detection module, cascaded behind the residual network assisting the recognition of pornography images. And our method makes a good performance based on the research work of pornographic image detection