Global Image Forensics Based On Statistical Features

In this era of technological manipulation,people can easily tamper with images due to the strong editing capability of image processing softwares,which greatly threatens the authenticity of images.Some malicious tamperers sometimes falsify images in order to hide or change important information,resulting in false information in the images,which greatly threatens the authenticity of the images.When these fake images are disseminated by various media means,there will inevitably be some confusion or misunderstanding in public cognition.In order to judge the authenticity of images,many researchers have proposed many effective algorithms for different types of image tampering to achieve effective forensics of digital images.The types of image tampering can be approximately distinguished into two categories:local tampering and global tampering.Among them,global tampering mainly affects the visual effect of the image,such as JPEG compression,filtering and contrast enhancement.In our work,the detection methods of histogram equalization and median filtering in global image tampering are studied.At present,the main problem of this kind of detection method is that it cannot guarantee the detection effect under the condition of small resolution image and compression post-processing.In order to solve the problems in the above global tampering detection methods,our work analyzes and studies the detection of histogram equalization and median filtering,and puts forward the following two schemes:1.For each input image,the method first calculates the adaptive grayscale range of feature extraction on the cumulative distribution function of the input image.Then,it uses the similarity between the cumulative distribution function and the identification function in the adaptive grayscale range,as well as the zero-value gap on the histogram,to form a feature vector.Finally,through the comparison of different classifiers,the method selects the classifier to classify.It can be seen from the experimental results that the algorithm in our work is suitable for the detection of conventional histogram equalization operations,and robust to the post-processing of JPEG compression.It also has a better detection effect on small-resolution images.2.Aiming at the median filtering in global tampering,an image median filtering detection method based on statistical features is proposed.The median filtering trace map is constructed by the structure tensor and median filtering residual.Then the classification features are extracted from the trace map by the transition probability matrix of Markov model.Finally the classification is performed by support vector machine.It can be seen from the experimental results that the algorithm is not only suitable for detecting median filtering operations of different intensities,but also suitable for small resolution images,and it is also robust to post-compression processing.