Image Sensitive Text Information Identification Based On Emotional Polarity Discrimination

With the evolution of Internet technology,information interactivity has been continuously improved,and its communication methods and channels have become increasingly diversified.The diversity of information dissemination methods to make people get rich data information,but also to the sensitive content information provides diversified way around,such as pornography,violence,and illicit,cults,and reaction,etc.,and brings great challenge to information security of the network information regulatory review work.As government departments strengthen the regulation of network information security,ordinary text sensitive information has been effectively curbed.many illegal organizations and individuals begin to release sensitive information in image way for online communication.Therefore,how to effectively identify the sensitive information of network image is an important research direction in the field of network information security supervision.Image sensitive information recognition can be divided into two categories:sensitive information recognition based on image text content and sensitive information recognition based on image visual content.This paper carries on the research with the image text content,the main research content is as follows:(1)Image character recognition technology: image character recognition is the key to sensitive to image character information recognition,due to the background image is divided into two categories,simple and complex,in the image preprocessing stage,the use of Gaussian filter and adaptive threshold binarization method to eliminate the image background noise,and then connect the content of the image after binarization area;For text area block extraction and text recognition,this paper selects two layers of restricted boltzmann machine(RBM)to distinguish and select the text area of the connected region.Using deep belief network(DBN)algorithm,the text is identified.(2)Multidimensional word web technology: because of the sensitive the evasion of the text,this paper puts forward the library based on sensitive keywords and emotional polarity thesaurus of multidimensional word net diffusion method,the character of each word to spread information toward the synonyms,words sound around,homophone,structure character,to update and improvement of sensitive keywords and emotional polarity of library.(3)Image sensitive text information identification based on emotional polarity discrimination: according to the text emotional polarity calculation method and sensitive image discriminant mechanism to identify sensitive text information,improve the identification accuracy of sensitive text information.In order to verify the validity of the relevant algorithms in this paper,the validityof the algorithm was validated for the manually collected data sets,and the relevant experimental results were compared with the results of other algorithms.Experimental results show that the proposed algorithm has higher recognition accuracy in text area discrimination and text recognition,and can better recognize image-sensitive text information.