Research On Passive Blind Forensics Of Copy-Move Tampering Digital Images

Today, we are witnessing the rapid development of advanced technology such as computer technology, digital technology and so on. Digital media has become the main medium of communication in information age. And a variety of digital images have become the most common carrier of information in daily life. With the rapidly developing of Internet and widespread popularizing of digital camera, meanwhile, powerful image processing and editing tools such as Photoshop obtain an unprecedented development. Ordinary users now can easily edit, beautify and even generate a digital image if they want. And these who are modified or generated by software can often mix truth and spurious with an astonishing degree. In recent years, a wide variety of news pictures frauds, academic fraud pictures, have exerted a great influence on people’s daily life and social stability. The problem of digital image forensics has become an urgent problem in the development of economy and society.Digital image tampering methods are various and copy-paste tampering is a representative tampering technique among them. This paper focuses on the research of the copy-paste forgery identification and authentication in the same image. The existing related methods include:exhaustive searching method, image autocorrelation forensic method, precisely matching method, method based on quantization and discrete cosine transform, method using the main transfer vector, method based on wavelet transform and singular value decomposition, etc. Although the methods above can complete copy-paste tampering identification and authentication, but most of them work with features of high dimensions, and the timeliness is low.Against to shortcomings of the above algorithms, an identifying and authenticating method of image copy-paste tampering based on DCT clustering is proposed. Relevant experiments are completed, and tests that anti JPEG compression, noise and blurring are carried out. In this paper, first of all, we divide the measuring image into overlapped blocks. Secondly, extract truncated DCT quantified features from each sub block. Then, take cluster analysis of the features using K-means clustering algorithm, and execute dictionary sorting for sub block features of each cluster. Finally match the similar sub blocks, and optimize the detection result with morphological filters. The experimental results show that, this method not only can realize image copy-paste tampering identification and authentication, but also has considerable robustness for common post-processing operations as the JPEG compressing, Gauss white noise adding, Gauss blurring.