Research On Passive Forensics For JPEG Image Tampering

With the rapid development of digital technology and the emergence of powerful imageprocessing software, it is getting easier to tamper with digital images without leaving anyobvious visual trace. While people are enjoying convenience and happiness which are broughtby those tools, the unintentional or intentional tampering of image appears. If the tampered orcounterfeited images are widely used in official media, scientific research, forensic evidence,etc, it may give rise to some big problems which potentially may have deep moral, ethical andlegal implications. Therefore, the authentication of digital images has been an importantresearch.In social life, JPEG format is popularly used in most digital cameras and also has beenapplied widely in the image storage and transmission on network, Therefore, analyzing thistype of image may play a useful role in image forensics. In this paper, the overview of presentresearch situation of passive forensics at home and abroad is studied; Especially, we havedetailly description about the existing forensic technology of JPEG image tampering; In thelight of classical algorithms, we proposed two new detecting schemes: On one hand, a blinddetection scheme is based on the JPEG block artifacts; On the other hand, we make use of thedouble quantization effect which hidden among the DCT coefficients for detecting andlocating the tampered areas. The main research innovations and contributions include thefollowing two aspects:1. According to the inconsistencies of quantization tables, a novel approach is proposed todetect JPEG image forgery. The primary quantization table of tempered image is firstlyestimated. Then the blocking artifact measure (BAM) of each8×8DCT coefficient block iscomputed using the estimated quantization matrix, and the BAMs of all blocks are sorted indescending order. Select a certain number of blocks of which the BAM is in the front of theorder. Labeling the connected components of blocks in order to eliminate some abnormalblocks and gain the connected doctored blocks. Experimental results demonstrate theproposed approach can effectively detect the tampered image whose first quantization step is greater than or less than second quantization step.2. The double quantization effect in JPEG images is becoming important clue in detectingimage forgeries. As many detection methods are based on the block artifacts of DCTcoefficients, a new blind detection scheme which makes use of the double quantization effectis proposed. By comparing two posteriori probabilities of untampered region, which arecomputed by histogram of DCT coefficients and interval length respectively, a feature thatcan distinguish the tampered block and untampered block is extracted, and then the featurevalue of each DCT block is calculated. After setting a threshold, the tampered blocks whosefeature values are larger than the threshold are obtained. Finally, the tampered region islocated by labeling the connected components. Experimental results demonstrate that theproposed scheme can detect and localize the tampered areas more accurately with respect topreviously proposed methods, especially for colorful or rich texture images.