Image Splicing Blind Detection Based On Enhanced Mavkov Model In DWT Domain

The development of image processing technology has made images,once the most reliable form for recording information, not reliable.Therefore, it is of great significance to study image forensics. There aremany forms of forgery operations to manipulate image content, of whichsplicing is the most general and basic one. Several methods have beenproposed to detect image splicing and they perform effectively for digitalforensics.This paper introduced the most representing splicing detectionmethod based on image natural statistical model, and then madeimprovement in the parts of image transform and Markov model. Imagesplicing oriented best wavelet sub-band selection and Markov stateselection method based on observed values’possibilities were proposed.For image transformation, the paper analyzed the reasons for DWT’ssuitability to image splicing with its theory and characteristic, and justifiedthat sub-bands in different scale and direction carried different splicinginformation and higher-order sub-band was most fit for splicinginformation extraction. For Markov model, the paper firstly analyzed theweakness for traditional Markov states selection method by roundingcoefficients and taking threshold, and then matched different functionmodels with the distribution and mapped coefficients to Markov statesbased on observed values’possibilities.At last, simulation experiment was conducted on Columbia imagesplicing detection evaluation database. Experiment results firstly justifiedthe theoretical analysis for the proposed blind splicing detection method.Then the extracted features were compared with other detection methods,and it proved to have the best detection accuracy, verifying the efficiency of our proposed method.