APPLICATION OF MAСHINE LEARNING METHODS FOR DETECTING OF JPEG IMAGE INTEGRITY VIOLATIONS

Subject of Research. The paper presents the study on the JPEG image integrity violations and existing methods of their detection. We propose a method for detection of modified image and the source of its modification. The method gives the possibility to determine the original image and camera model that recorded it. Method. The method was developed with the use of machine learning tools. The following machine learning methods have been studied: naive Bayesian classifier, decision tree, logistic regression, k-nearest neighbors, SVC, random forest. The base for model training was formed by the original photos from website www.steves-digicams.comthat were modified by different graphic editors. The proposed method uses JPEG-image structure in byte view, namely, markers. Availability of markers and their number were suggested as classification features. Main Results. The trained model has demonstrated high classification result equal to more than 95%. Among all evaluated algorithms the two ones have shown the best results: decision tree and random forest. Decision tree was chosen as the best one upon stability criterion. Practical Relevance. Thereceived result can be practically applicable in the area of forensics and information security.

1. Khatuntsev N.A., Lizorkin A.M. Method of proof invariably images in the computer-technical expertise (from expert practice). Theory and Practice of Forensic Science, 2014, no. 3, pp. 69–73. (In Russian)
2. Farid H. Digital Image Ballistics from JPEG Quantization: A Followup Study. Technical ReportTR2008-638.Dartmouth College, 2008,6 p.
3. Photo Tampering throughout History. Available at: http://pth.izitru.com/1994_02_00.html (accessed: 20.10.2017).
4. Shcherbakov S. Metadata in Digital Photography. 2005. Available at: www.ixbt.com/digimage/metadxph.shtml (accessed: 20.10.2017).
5. Fedotov N.N. Forensic - Computer Criminology. Moscow, Yuridicheskii Mir Publ., 2007, 432 p. (In Russian)
6. Beck T. How to Detect Image Manipulations? 2017. Available at: headt.eu/detect-image-manipulations (accessed: 30.10.2017).
7. Color Adjustment: HSV colorspace. Available at: fotoforensics.com/tutorial-coloradjustment.php (accessed: 30.10.2017).
8. Recommendation T.81. Information technology – digital compression and coding of continuous-tone still images. Part 1. Requirements and guidelines. CCITT, 1993, 186 p.
9. Vatolin D.S. Image Compression Algorithms. Moscow, MSU Publ., 1999, 76 p. (In Russian)
10. Shelepov M.I. History of creation, organisation, structure and use of JPEG graphic format. Available at: www.kolpinkurs.ru/stati/jpeg.htm (accessed: 25.10.2017).
11. Domingos P. A few useful things to know about machine learning. Communications of the ACM, 2012, vol. 55, no. 10, pp. 78–87. doi: 10.1145/2347736.2347755
12. Pedregosa F., Varoquaux G., Gramfort A. et al. Scikit-learn: machine learning in python. Journal of Machine Learning Research, 2011, vol. 12, pp. 2825–2830.
13. Cai Y.I., Ji D., Cai D.F. A KNN research paper classification method based on shared nearest neighbor. Proc. NTCIR-8 Workshop Meeting. Tokyo, 2010, 5 p.
14. Quinlan J.R. Induction of decision trees. Machine Learning, 1986, vol. 1, no. 1, pp. 81–106. doi: 10.1023/A:1022643204877
15. Hsu C.W., Chang C.C., Lin C.J.A Practical Guide to Support Vector Classification. 2016. URL: www.csie.ntu.edu.tw/~cjlin/papers/guide/guide.pdf (accessed: 25.10.2017).
16. Ali J., Khan R., Ahmad N., Maqsood I. Random forests and decision trees. IJCSI International Journal of Computer Science, 2012, vol. 9, no. 5, pp. 272–278.

This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License