Research On Image Hash Method For Content Forensics

With the rapid development of image processing technology and the popularity of digital products,multimedia data such as digital images and videos have been widely used because of their accessibility and real-time performance.However,many powerful image processing software make it easy to edit and modify image content.If some maliciously tampered images are used in the fields of politics,military affairs,medicine,scientific research,news media and judicial forensics,it will greatly reduce the credibility of image information and cause serious negative impact.Therefore,it is significant to research and develop digital image tampering detection technology and make it to be a tool to maintain the order of modern digital society.In this work,we study the image hash technology and hash-based image content forensics,and the main works are as follows:We studied the image hash technology based on content forensics,and propose a lightweight perceptual image hash generation method and a hash-based fine-grained image tampering detection method.In the proposed method,we take the gray mean of image bit-plane as the global feature,take the position of the image vertices as the layout feature,take the color layout descriptor of the inscribed rectangle of the super-pixel as the local feature,and use these features to generate image hash code.We use the proposed image hash to construct an image tampering detection and fine-grained tampering localization algorithm.In the proposed method,the global features are used to detect if an image has experienced tampering attacks,layout features are used to correct geometric transformations,and local features are used to detect the location and shape of the tampering regions.This method has compact hash length,high tampering detection and tampering localization accuracy,and can detect different types of forgery attacks,for example,an image simultaneous suffer splicing and copy-move attacks.The experimental results show that the proposed method is provided with pixel-level detection accuracy and stronger robustness to tolerating content-preserving manipulations and geometric distortion,and it can achieve a better trade-off between perceptual robustness and perceptual discrimination.Compared with the existing methods,this method has greater detection performance.Aiming at the contradiction between the length of hash code and the detection accuracy for tampering regions,we propose a perceptual image hash method that is provided with more compact hash length.In the proposed method,we use the Hu moment feature,layout feature,local color moment feature and pixel intensity statistical feature of the image to generate hash code.The tampering localization algorithm consists of two stages:in the first stage,we divide the image into regular blocks,and use the color moment feature and pixel intensity statistical feature of the image blocks to perform coarse-grained tampering localization;in the second stage,we use super-pixel segmentation algorithm to smooth the edge of coarse-grained tampering localization results and obtain fine-grained tampering localization results.This method is provided with more compact hash length,higher tampering detection and tampering localization accuracy,and it can effectively detect the shape and location of multiple tampering regions.This method has stronger robustness,it is provided with a good trade-off between robustness and sensitivity,and it has higher computational efficiency.