Fast Image Forgery Detection Based On Block Matching

With the progresses of computer technology and the widespread use of Internet technology,it is very easy to tamper with sounds,images,and videos.It is also easier and easier to spread fake sounds,images,and videos on a large scale.On the one hand,digital counterfeiting technology is becoming more and more mature,and on the other hand,to accuratly discrimination between real and counterfeit content requires certain expertise.At present,the digital forensics technology obviously lags behind the digital forgery technology,which has drawn great attention from researchers.This thesis focuses on the copy move forgery detection.For the current copy move forgery detection algorithms,block-based,key-point-based,and segmentation-based detection methods are mainly used.These forgery detection methods cannot guarantee good effect and low time complexity under various attacks.Aiming at the above-mentioned problem of time complexity of image forgery detection,an in-depth study is performed on the basis of block matching-based image forgery detection algorithms.The main research contents include the following two aspects:1.Based on the idea of matching between block classes,a fast copy move forgery detection algorithm based on grouped SIFT is proposed.First,the image is divided into blocks,and each block is classified.Then,feature points are extracted as block features in each image block.Finally,Finally,block matching is performed by matching between classes to locate the forgery area.According to the analysis of the experimental results,this method has better detection effect and lower time complexity through inter-class matching and feature points matching.2.Based on the design idea of image block matching,a fast copy move forgery detection algorithm for multi-scale image block matching is proposed.The algorithm performs multiscale spatial modeling on Patch-Match.The core idea of Patch-Match is Approximate Nearest Neighbor Searching.Approximate nearest neighbor search has always been an important method of data query in computer vision and big data applications.In this paper,we combine the idea of approximate nearest neighbor search with forgery characteristics of Copy Move,and propose an efficient and fast image feature matching screening method to reduce the feature matching screening time of previous detection methods.The experimental results show that this algorithm applies the advantages of multi-scale Patch-Match,which effectively reduces the time complexity and the detection effect is better.