| With the development of Internet,network streaming technologies and multimedia information retrieval,images and videos have become the mainstream carrier of information exchanging and multimedia service.With the emergence of the 4th generation(4G)digital communication network,information dissemination of instant communication has been more rapid and extensive,like WeChat.While it is convenience to people with internet,content security has also become a more thorny issue.The dissemination of sensitive information,such as political reactions,pornography,commercial secrets and personal privacy,has a greater impact on national security,social stability and the healthy growth of adolescents.Currently,filtering sensitive information in plain text has been relatively mature,but there are still some challenges in the extraction of text superimposed in images,and rapid identification of sensitive images and videos.For content security of network monitoring,Firstly,we analyze the embedded text and image content in the Internet.Then,In order to secure the sensitive data transmission and storage,we need to study the encoding and decoding algorithms of private protocols for text and image compression.Finally,the text and image information are recognized to realize the sensitive content monitoring of mobile Internet multimedia.Our main research achievements are summarized as follows:1、We propose a new image text region location method based on Gabor texture and neural networks,which achieves good results on multilingual text detetion.Text regions in images are hard to be detected because of the complex background.Using the texture directions of the text region,we propose a new technique on text region location in images based on Gabor character and neural networks classifier.First,the direction textures of original images are described by Gabor features with several directions and scales.Then,the BP neural networks classifier is trained based on the Gabor features of text regions and the non-text regions,which can be used to detect the text regions.Some experimental results illustrate that our algorithm greatly improves the accuracy and robustness than other methods.The proposed algorithm is suitable to locate not only Chinese text but also the text in English and other langues.2、To extract the embedded text in images,a fast text-embedded image text extractionalgorithm based on homogeneous space mapping is proposed,which achieves 86% on recognition rate.Text-embedded images are popular in the mobile Internet to spread malicious information.A fast text-embedded image Chinese text extracting algorithm based on homogeneous space mapping is proposed.Image enhancement functions are used to highlight the edge and texture features of images.Sobel operator is used to extract the edge feature and wavelet packet is used to extract the 24-dimensional texture feature vectors in the enhanced images.The texture features and edge features are used to describe the homogeneity of an image,which construct the homogeneous feature map of the image.The differences between the non-text and the text region-homogeneity are used to distinguish them and reduce non-text regions further.Thus,the text regions are highlighted.Then,homogeneous text samples are used to train the text region detector,which greatly reduce the computational complexity.Finally,the characters are segmented and recognized.Some experiments to verify the validity and practicability of the proposed algorithm have been conducted.The recognition rate achieves 86%,which is higher than that of other methods in industry.The algorithm is verified on the operator’s malicious information monitoring system,which provides secure malicious filtering performance.3、To solve the compression and transmission problems on sensitive text data,we propose a novel parameter local optimization algorithm for text data compression based on Glomb coding,which is superior over Winzip in compression performance.In view of the political class sensitive text data,the security of data in storage,transmission and application is a difficult problem.The private protocol data decoding algorithm while keeping the data compression efficiency provides the necessary security.A novel algorithm for text data compression is proposed based on parameter local optimization.The global parameters optimization is converted into the local parameter optimization,then Glomb code principle is used to solve the local optimization,and a parameter inferencing method is derived.A modified LZ77 scanning algorithm is proposed,in which a new prediction code is developed and the prediction data is preserved.The parameters such as prediction data,offset,match data length and preserved text data are loaded into MQ codec,in which the data can be compressed.To improve the compression efficiency,the corresponding compression algorithms and the context design algorithm are proposed.The experimental results show that our compression algorithm has an advantage in compression effect over the Winzip on the data such as texts,word documents,Clanguage program codes and images.Compared with WinRar,our algorithm achieved almost the same compression results for texts,word documents,C language program codes except images.4、To solve the compression and transmission problems on sensitive data,A novel algorithm for high precision image compression based on error optimization coding is proposed,which achieves good compression effects.In order to protect the sensitive data in Internet image transmission and storage process safety,it is necessary to study the private compression algorithm to guarantee the data security,which should achieve high precision image compression efficiency,while ensuring the subjective quality of the reconstructed image.A better subjective quality for reconstruction image is obtained.A novel algorithm for high precision image compression based on error optimization coding,where the errors produced in JPEG-LS runlenth coding process are quantified,and the quantified value is analyzed in which the correlation is removed,and the analyzed data is encoded in MQ.The inverse quantified value is done in terms of the quantization gap,which is added with the value reconstructed by JPEG-LS decoder.Finally,the added data is filtered.The algorithm is analyzed in theory,the simulation is carried out,and the simulation results are analyzed,and the performance is compared with the one by JPEG2000.The results show that the proposed algorithm can compress the high Precision Image data with high efficiency and the information loss is less than the one produced by normal compression algorithms.Although the objective quality is lower than the one produced by JPEG2000,the subjective quality is better.5、To solve the sensitive text and video filtering problems in 4G communication,a sensitive video and image recognition system based on text and face recognition is developed,which is applied to guangdong mobile network systemFor mobile Internet applications requirements,it is necessary to study the fast identification application system based on sensitive text and video.The research focuses on the sensitive text and video filtering technologies in 4G communication.It mainly includes politically sensitive texts and pornographic images.The key frame of the video is the synopsis of the video,and the computation is reduced only for sensitivity to key frames.For the recognition of pornographic images,human skin characteristics are very important clues.Human skin color has a certain clustering property in the color space,so it is more practical to use skin detection in complex background.The sensitivity of multimedia isevaluated by combining the features of human face and human organs,and tracking sensitive key frames to judge the sensitivity of the whole video.It provides security for mobile communication content security on the Internet pipeline. |
PDF Full Text Request