The Research On Image Passive Forensics Based On Feature

With the popularization of multimedia technology,digital image forgery is becoming more and more frequent,which brings great negative influence to the society.There is a pressing need for a new technology to resist the many losses caused by this behavior,digital image passive forensics technology has also been the academic attention.Among the many image tampering methods,copy-paste tampering is one of the commonly used tampering methods.The tampering is only to copy a region in the image and paste it to the same or different images to be tampered with the region,and finally to tamper with the region fuzzy,noise reduction and other processing,the whole process is simple and easy to tamper with the results from visually deceive the observer.for such tampering,there are a lot of published literature to propose their own detection scheme,but the detection ability of these schemes in image smooth regions(such as the sky,snow,etc.)has yet to be improved.This paper aims at this problem from the following three aspects:1)A copy-paste tamper detection algorithm based on Zernike moments is proposed.The algorithm mainly uses the principle that the tampering area and the offset angle of the reference area of the copied area are consistent,and then the main transfer vector is used to determine the tampering area.When the tampering region is the same as the copied region,the algorithm can detect one or more tampering regions,but the tampering action will also do some edge processing in the tampering region,and additional operations such as noise reduction will make the image more realistic.The detection ability of the algorithm will be seriously degraded,and the generation of this problem paves the way for the exploration of another algorithm in this paper.2)A copy-paste tamper detection algorithm based on WLD feature is proposed.The algorithm firstly uses the direction angle component to quantize the gray image of the natural image,and then overlapping blocks for the image.Then the binary tree is traversed and matched.Finally,the tamper detection result is found by threshold control.In the process of matching algorithm,the dictionary is used to find the starting block of each suspected tampering region(that is,the root of the binary tree),which effectively reduces the time complexity of the whole algorithm.Threshold control by multiple sets of parameters,for different tampering images,can find a most satisfactory test results.3)Using the CASIA v2.0 library provided by the Chinese Academy of Sciences,we compare the algorithms based on Zernike moments,the algorithms based on WLD features and an algorithm based on SIFT feature points.The final experimental data show that based on WLD features Of the algorithm has a better performance.Further experiments show that the algorithm based on WLD features has good robustness to Gaussian blur and gamma correction.This series of experiments also proved that the WLD-based algorithm has better detection ability in the scene with smoother or high degree of blur images.