Research Of Image Passive Forensic Algorithm For Detecting Removal Object By Block Padding

In recent years,with the increasing popularity of image editing software,images can be modified easily.Some people with bad intentions would forge the image with ulterior motives,and the old saying that "seeing is believing" is no longer applicable.Therefore,it is imperative to find reliable and effective forensics methods to authenticate the image.The difference between passive forensics and active forensics is that passive forensics does not need to add protection information in advance.Instead,it identifies images by tracing the clues left in the process of forgery.Therefore,passive forensics has attracted more and more attention of researchers in recent years.Among various forgeries,removing an object in the image,moving its position away from its original position,or adding a new target in the image can easily affect the people's intuitive judgment of the image.Since the block-based image inpainting can also be used as a target removal method,the research on forensics based on block filling has high practical value.This thesis aims at the object removal based on block filling and studies the detection techniques of such tampered measures.First,this thesis analyzes the performance of existing forensic algorithms against block padding tampering and discusses the influence of two important parameters involved in the algorithm : the similarity and the length of filter vector on the recall rate and the precision.The result indicates that since the Euclidean distance between each pair of repaired blocks is relatively long,when the length of filter vector is 2,the similar blocks in the background area can be effectively filtered without any missed detection,no matter whether the removed target is in a flat background area or a more complex texture area.Moreover,it is unreasonable that the existing forensic algorithms sets 12 as a similarity threshold,resulting in a certain degree of missed detection,this thesis improves the similarity threshold.In order to verify the detecting performance of different inpainting algorithms,this thesis adopts three classic inpainting algorithms to remove the target.The experimental result demonstrates that,after improving the similarity threshold,the recall rate and the precision of the algorithm in both single-target and multi-target scenarios have been improved for the one-to-one block padding tampering,where the recall rate has been improved reached 5%.Second,we aim at the problem that the larger segmentation and higher time complexity is pointed out in the previous analysis,a fast forensic algorithm based on small neighborhood is designed,which can improve the calculation speed of the algorithm while keeping the recall rate and the precision at a high level.This algorithm uses smallneighborhood(complete 8 neighborhoods)to detect suspicious tampered pixels and reduce the amount of calculations.In the process of locating the tampered region,this algorithm proposes a regional overall labeling method to distinguish the tampered area from the reference area.Experimental results show that time efficiency is significantly improved on the basis of ensuring high recall and precision.