Research On Image Active Authentication Algorithm Based On Content Perceptio

With the popularity of image editing tools,the cost of image tampering has dropped sharply,and unscrupulous elements can easily use tampered images to commit criminal acts,posing a great threat to social stability and national security.There is an urgent need to design effective image content authentication algorithms to curb malicious tampering in social media.Current positive authentication algorithms are mainly divided into precise authentication algorithms and content-aware authentication algorithms,which perform authenticity verification by extracting pre-embedded/saved information.Accurate authentication algorithms are able to protect image integrity well,and the authenticity is considered to be corrupted when an image is modified.However,with the popularity of social networks,images often undergo unintentional modification operations such as compression and quantization during transmission,and the precise content authentication algorithm will not be suitable for content authentication of social images;the semi-fragile watermarking algorithm with content perception and the perceptual hashing algorithm are able to extract watermarks normally for a certain degree of unintentionally modified images,and both distinguishable for maliciously modified images,providing a viable solution for social image content authentication.Existing content authentication algorithms with content perception still face a number of challenges.Firstly,there are various image operations that may be encountered during image transmission,and content authentication algorithms need to be robust to noise,compression,quantization and other operations while accurately distinguishing tampered images,which poses higher requirements for algorithm robustness design;secondly,there are various image tampering operations,and the similarity between small region tampered images and real images is very high,and it is difficult for current algorithms to explore the difference between small region It is difficult for current algorithms to explore the differences between small region tampering and real images.This dissertation aims to investigate a content-aware image authentication algorithm based on deep learning techniques.By leveraging the powerful fitting ability of deep learning,the algorithm is able to maintain robustness against unintentional modifications while distinguishing maliciously tampered images.The research content is as follows.(1)For the problem of lack of robustness,this dissertation proposes a deep semi-fragile watermarking-based image content authentication scheme.Images often undergo compression and noise operations during transmission,which will have an impact on the image content but will not destroy the authenticity of the image semantics.This proposal proposes to study image authenticity authentication based on deep semi-fragile watermarking,simulating the unintentional modification operations such as noise and compression and malicious tampering operations such as copy-paste and splicing experienced during image transmission,and using the powerful fitting ability of deep learning to balance the robustness of semi-fragile watermarking to unintentionally modified images and the vulnerability to malicious tampering images to achieve accurate authenticity authentication.(2)To address the weak tampering perception problem,this dissertation proposes a depthaware hash-based image content authentication scheme.The semi-fragile watermarking scheme requires prior embedding of authentication information in the image for authenticity authentication,which will lead to a certain degree of image distortion.This scheme proposes to study the deep-aware hashing algorithm,design an end-to-end perceptual hash generation network for hash extraction of images,constrain the separation of tampered and real image hashes from each other while ensuring that the perceptually similar image hashes are similar,and use the feature consistency analysis module to enhance the scheme’s ability to perceive weakly tampered regions,so as to authenticate the authenticity of images without modifying the images themselves.