Research On Several Algorithms For Digital Image Authentication

Due to the rapid development of science and technology, a lot of tampered images have been appearing in our daily life. It is difficult to distinguish the truth of them for us. Therefore, it is important to develop an image authentication technology to accurately certificate the Authenticity of images. At present, there are two main kinds of techniques for image authentication: one is active image authentication technology which includes digital watermarking technology and image perceptual image hash technology, and the other is passive authentication which is referred to as passive-blind image forensics.Image perceptual hash technology relies on prior information such as the pre-extracted content signature. The main contents for it is the conflict that hash value is supposed to robust to common content-preserving manipulations (CPM), such as lossy compression, while sensitive to malicious tamper. To overcome the contradictory requirements, a two-step general robust hash which is composed of a global hash and a local hash can meet the two important features, But them don't enough sensitive to malicious tamper which need to be further researched. How to make the tamper detection and localization still work well under some typical CPM is another open issue. To overcome the problem a tampering localization model (TLM) based on machine learning is proposed. However, whether the training samples of the model or CPM is generic is o be certified. In the passive authentication field, How to detect copy-paste of JPEG image by passive-blind image forensics technology need to be further researched .In this paper, we'll improve the existing algorithms to make them certificate the authenticity of the image more effectively than the original ones.The main research and innovation are as follows:Firstly, we research the robust hash algorithm based on structure information. Based on it, we improve the matching algorithm, and in the process of getting local hash, unlike the original method, we consider not only the mean information but also the variance information. The experimental results show that the improved algorithm is better than the original one in authentication granularity , tamper detection precision and recall, but the robustness is the same as it.Secondly, we research on TLM and design experiments to prove that the training samples of the model or CPM is generic.Thirdly, we research on DWT based robust image hashing algorithm, and use TLM to improve it. Experimental results show that the precision of tampering localization of improved algorithm is high than the original algorithm under some typical CPM. Finally, we research on an passive authentication algorithms based on JPEG artifact grid extraction and extend its range of available evidence. Experiments demonstrate that the algorithm can detect the copy area effectively as well as the bmp format copied image, what's more it is more effectively detected than the copied image which is JPEG compressed.