Research On Pornographic Image Recognition Based On Deep Learning

With the rapid development of the Internet,the spread of network pornography is more and more rampant,which has a very bad impact on the life and learning of Chinese netizens,especially the minor netizens.The state spare no effort to crack down on the illegal acts of Internet dissemination of obscene pornography,and the relevant laws and regulations are constantly improving.At the same time,it is necessary to filter and block the pornographic information on the Internet.Because of its large scale,wide spread and great harm,network pornographic pictures are one of the targets that we focus on to clean up the network environment.With the arrival of 5G era,the mobile Internet with large bandwidth further intensifies the spread of digital pictures,the scale of pornographic pictures is also growing rapidly,and the performance requirements of the recognition algorithm for bad pictures are also increasing.This paper summarizes the existing pornographic image recognition algorithms,mainly analyzes the traditional recognition algorithm based on the region of interest and the recognition algorithm based on deep learning,summarizes the technical characteristics of the two main uses,compares their advantages and disadvantages and the scope of application.Because pornographic images have the characteristics of massive scale and variety,the traditional image recognition and filtering methods have some limitations on the prior features,which affects the recognition accuracy of the algorithm.However,the pornographic image recognition algorithm based on deep learning has no limitation of artificial feature selection.It has more expansibility in large-scale sample learning and feature extraction,and can meet the target requirements of pornographic image recognition.Therefore,after summarizing and analyzing the existing image filtering technology,this paper focuses on the image recognition algorithm based on deep learning,proposes a pornographic image recognition algorithm framework based on R-FCN and Res Net,and makes two improvements on the basis of the existing network framework,mainly on the Ro I pooling layer and location feature extraction of the network.In order to fully verify the effectiveness of the algorithm,this paper uses a large-scale data set containing more than 20,000 pornographic images for learning and training.In the experiment,the main task is to identify the exposed sensitive parts of the human body in the images.Once the images are detected,any exposed sensitive parts of the human body are marked as pornographic images for filtering.Experiments show that the accuracy of the algorithm is 96.2% for pornographic images and only 2.2% for normal images,which is better than the traditional method.At the same time,in order to verify the effect of the network architecture,the same group of comparative experiments are carried out in this paper.The results show that the deep learning neural network architecture and parameter setting verified in this paper can significantly improve the accuracy of pornographic image recognition.