Research On The Methods Of Statistical Feature Based Image Retouching Detection And Image Recapturing Detection

With the development of technology and the wide application of image editing software,image editing and altering have become very easy.It makes people pleased and enriches people's visual experience.However,it has caused extremely serious negative impact in many fields,such as government affairs,finance,law,news media,science and technology,medical field,etc.In recent years,with the frequent occurrence of image tampering events,digital image forensics has become an important topic in academic research.In this paper,our main focus is on the research of image retouch forensics and image recapturing detection.The major content can be outlined as follows:Aiming at the operations of current popular Meitu software,such as Meitu xiu,Adobe Photoshop,Corel PhotoPaint,PotraitPro Studio,etc,we propose an image retouching detection method based on statistical features.Considering that image retouching operation would cause the changes in texture details,color and brightness,etc,we comprehensively utilize the Local Binary Pattern(LBP)which can reflect the image local texture details,the digital features of gray level co-occurrence matrix(GLCM),the tamura texture features,as well as the mean,standard deviation,and skewness of four image channels which can reflect color and brightness of the image,to construct distinguishable statistical features,and then use the support vector machine classifier to distinguish the original image from the retouched image.Through a large number of simulation experiments,the results show that the detection accuracy of the proposed method reached 96.86%,and has good robustness.Compared with other similar algorithms,our method has better performance.We proposed a method for image recapturing detection based on multi-feature fusion.Unlike the existing methods that used higher-dimension features followed by a classifiers,our approach is provide with low dimensional features and discard the classifier.We analyze the difference between the original images and the recaptured ones that is caused by the specular reflection and the high frequency wavelet components introduced in image recapturing process,and use the gradient of image specular reflection,the vertical high frequency component of wavelet decomposition,and the statistical magnitude of the beltrami flow to construct a representative features.Then we construct a discriminant model by analyzing the differentiability of these features,and combine with a threshold method to identify recaptured images.A large number of experiments have been conducted on the existing public recapturing image database,the experimental results show that our algorithm has better detection rate in spite of low feature dimension.For content-preserving image processing operations,such as JPEG compression,gamma correction,noise addition,filtering and so on,the algorithm has better robustness.