Study On The Technology Of Color Digital Image Synthesis Tampering Blind Forensics

In recent years, with the development of electronic, computer and otherinformation technology and the image recording devices become more popularizationsuch as digital cameras, digital imaging has been widely seen in the people’s daily lifeand work, becoming an important way for people to get and delivery information. Atthe same time, rapid development of image processing and editing software alsoallows ordinary users more easy to use image editing tools to modify or synthetizedigital image, and then according to the wishes of individuals, commercial purposesor to achieve the effect of media hype build entrusts its digital images which is forged,this will be subverting the traditional vision "seeing is believing" concepts and gaverise to public crisis of confidence, coupled with the recent popularity of mobileintelligent Terminal, micro blogging such as social networks to promote, makeinformative, intuitive, and highly compelling digital images spread more quickly andspread dissemination are deep and broad in scope. Thus, on digital media, includingdigital image information authenticity and integrity validation to maintain healthy andsustainable development of the information industry as one of the key problems to besolved urgently. At the same time, to guarantee public order, fight against malicious orillegal modification of trust image thereby maintaining judicial impartiality is also ofgreat significance.This paper begins with an analysis of the existing principles and relatedcharacteristics of blind detection of tampering image with evidence, and then for colordigital image blind tampered with evidence presented two tamper detectionalgorithms are as follows:Firstly，Between different camera to take digital photos of "copy-paste" tamperedwith, proposed " digital image forensics algorithm based on artificial neural networks".The algorithm by digital image brought interpolation characteristics to determinewhether an image has been tampered. This paper put the original image data inaccordance with the Bayer format divided into input data and expected output data assample data input BP neural network training, and then the image color information input trained neural network to predict the image obtained by the neural networkforecasting color information to determine the size of the error of the actual imagecolor information forged region. In order to further improve the accuracy of forensics,this paper proposes the use of adaptive threshold to determine the tampered area andrandomly selected image data by several training iterations to filter theuncontaminated sample data to train the neural network, on the one hand, fewer thecomputational complexity of the algorithm, the more important is to increase thedegree of approximation of the original interpolation algorithm excluded data ispollution.Secondly, for an image of whose partial image is copy-paste from another partof the image in order to hide an important goal or increase certain attributes, inprevious studies on the basis of the tampering detection algorithm proposed purequaternion SIFT feature extraction combining the algorithm by flash SIFT featureextraction and matching the extracted suspected tampered area, some of the keymatching point, and then coarse estimation tampered region, reuse the Quaternionsuspected tampering area image modeling phase tamper detection directly in hypercomplex domain, the algorithm reduces the color image processing, you must first beconverted to gray scale image chroma information loss brought about the possibilityof erroneous matching, thereby improving the accuracy of the algorithm.