Digital Image Forensics Technology Research

Benefited from the development of LSIC and information technologies, The digital camera as swallows from distinguished families fly into commoners' courtyards.Now, in our daily life, digital cameras have been almost completely replaced the traditional film camera become the main tools of video recording. However, compared with the photographs taken by traditional film camera digital photographs were more easily to been tampered.So, how to determining the picture have not been tampered before it been used had become a problem which news, economic and military, political and other areas must to facing. Traditional practices is embedded digital watermarks in photographs,once someone tampered the images it would undermine the watermarks,by this method we can judge whether the image has been tampered with. But this method would undermine the visual of the photographs,its not an ideal method. If we can find a way that not need to use watermarks but only use the photographs itself to determining whether the image has been tampered,it would be a perfect method.This paper studies the content can be divided into two parts,one proposed"Blind detection algorithms for forged images based on artificial neural networks". another to put forward an improved method of"Blind detection algorithms for forged images based on lexicographically ordered feature vectors".Firstly, according to the principle of digital camera put forward"Blind detection algorithms for forged images based on artificial neural networks". Pointed out, using the interpolated algorithm information which in images themselves to determine whether one image has been tampered with. If image contains two or more interpolated algorithm information, and that the image is being manipulated.And in this image, the small part which contains different interpolated algorithm information from other parts was the tampered region.To achieve this end, we use original image to training bp nerve network and fitting out the interpolated function of the image. Then inputing the original image to this interpolated function. Using the D-value with original image and the output of interpolated function to determining tampered region. To further improve the accuracy of determining, we put forward a method that using weighted value to find an subimage which had best interpolated relations. Using this subimage to training bp nerve network, also enhance the robustness of the pepper salt noise.Secondly, In first part of this paper we proposed a method to determining forged images which consisting of different digital camera taken.But in practical applications someone may forge images by two or more image which taken by one digital camera. People has put forward some method to determining this kind of forging,"Blind detection algorithms for forged images based on lexicographically ordered feature vectors"is one of them.In practical applications, image may have been fuzzy processed. So,similar image of the dictionary sorting may not be adjacent, and the algorithm would get false results. This paper put forward an improved method of this algorithm, using the distance of feature vectors as a parameter to ordering the feature vectors, then grouping dct coefficients matrix according to the frequency of feature vectors. This method improved the accuracy of the algorithm.