The Research On Anti-forensic Technology Of Digital Image

At present, most image forensics can detected the tampering operation without specialanti-forensic operation. However, for the profession interpolator with a good knowledge ofdigital signal process, the normal forensic technology can be easily fooled. In a word, thestudy of the anti-forensic technology can enhance the security and robustness of the forensictechnology. In this paper, we proposed the background of the anti-forensic and the researchsituation at home and abroad. Then we have a deep study of some classical anti-forensictechnologies.Firstly, this paper improves the previous JPEG compression anti-forensic algorithm，enhance the secrecy of the tampering operation at a low image quality cost. By graduallyraising the image compression quality factor and adding the anti-forensic dither into thetampering image, this paper removes the quantization artifacts among DCT coefficients.Meanwhile, we uses the minimum total variation denoising method retain the edge andtexture information. Experiments show the anti-forensic algorithm is more secretive and keepa better image quality than other algorithms. This means, the improved anti-forensic canremove the quantization artifacts without no more forensic artifacts.Secondly, the paper hiding the original source trace of the tampering image during thefingerprint-copy attack. By building the image imaging model in the source device andimprove the fingerprint-copy attack, it controls the effect of the true original source’s sensorpattern, on the other hand it adjusts the correlation of the true or fake original devices throughmethods like least square method. Experiments show in the proposed algorithm, thecorrelation with fake original source is far more high than the true original. This means thisalgorithm can have a fingerprint-copy attack without leaving the true original trace.In the last, this paper attacks the triangle test forensic. We calculate the variances ofrelevant parameters among every stolen images’ pixels, choose the bigger variance pixelswhich have a strong imaging sensor artifacts and abandon the smaller pixels which have aweak imaging sensor artifacts. By the chosen pixels, we construct an objective fingerprint inan incomplete matrix. Experiments show the algorithm could keep a high correlation withfake source and reduce the success rate of the triangle test. After improving, we could have afingerprint-copy attack while defending the triangle test.