Research Of Classification Algorithms For Blind Digital Image Forensics

The development of digital imaging technology and software brings people tremendousconvenience in image acquisition and processing. Meanwhile, a lot of negative problems suchas image forgery, steganography, etc are emerging. The blind digital image forensics isconsidered as one of the pivotal and efficient technologies to solve these problems. Therefore,carrying out the research on it has important significance for crime-controlling, defendingjudicial justice, ensuring public trust, and so on.The blind digital image forensics has two main interests, namely authenticity detectionand source identification. This paper carries out the research on source camera identification.Following the tow main approaches for digital image source forensics, pattern classificationbased on image features and correlation detection based on pattern noise, the advantages anddisadvantages of classical source camera identification algorithms are analyzed first, and thenthe improved algorithms are proposed. The main work can be summarized as follows:1. The whole imaging process is introduced, and the characteristics which areintroduced by the components of camera to the final output image are discussed in detail.According to the classical algorithms for source forensics, five feature sets are extracted fromimage to reflect the imaging characteristics. Through the simulation and comparison, theadvantages and disadvantages in performance of source camera identification for each featureset are analyzed.2. For each feature set can not reflect the imaging characteristics completely, animproved image feature based method for source forensics is proposed. The algorithmsynthesizes multi-feature sets to form a primitive feature set, and a sequential floating forwardselection algorithm is employed to do feature optimization to remove the redundant featuresand improve the capability of classifier. The experimental results show that the proposedmethod not only promotes the accuracy for source camera identification, but also improvesJPEG compression robustness.3. In allusion to the deficiencies of traditional pattern noise extraction method, a newpattern noise based algorithm for source camera identification is proposed. The noise varianceof reference pattern noise is analyzed first, and the interpolation pixels in reference patternnoise are abandoned. Then the fuzzy clustering algorithm is implemented on test image to dotexture complexity analysis, the regions with high texture in residual noise image are removed.The experiments show that the proposed algorithm has better source identification ability.