Blind Digital Image Forensics Research Based On Moment Features

Along with the rapid development of network and multimedia, digital image has become an important carrier of information. However, with the film image, digital image based on digital form can be tampered extremely easy. If these tampered images are used in scientific research, news reports, judicial evidence, and so on, it is bound to have bad impact to social stability. So, it is urgent to identify the authenticity of digital image. At present, there are two image forensic techniques: active forensic technique and passive forensic technique. Active forensic technique need to be embedded additional information such as watermark, which is of great limitations. Passive forensic technique, also called blind forensic technique, authenticates integrity and authenticity of the image based on the attribute information on itself not any other embedded additional information. This paper studied the theory basis and generated process of using moment features distinguish natural image, splicing image and computer generated image, and researched image splicing detection technique and computer generated image distinguishing technique combined with pattern recognition. Results obtained are as follows.Firstly, this paper respectively analyzes statistical characteristics of natural image, splicing image, and computer generated image to provide the theoretical foundation for detecting splicing image and distinguishing computer generated image.Secondly, in this paper, we introduced in detail the related theory and generating process of statistic moment features, and proposed an image blind forensics framework based on the moment features and pattern recognition.Thirdly, we studied the image splicing way that is the most common way of image tamper. By comparing the statistic characteristics of the natural image and splicing image, we proposed an image detection technique based on the moment features and image quality metrics and reduced the characteristic dimensions which effectively reduce the test time of authenticity. Experimental results show that this method has sufficient theoretical basis and outstanding detection effect. Fourthly, the identification technique distinguishing real image from computer generated image was studied and introduced moment features which can distinguish real image and computer generated image. Experimental results show that this method has good accuracy and strong adaptability.