Digital Image Forensics Based On Mobile Phone Camera

This paper mainly studies the method based on pattern noise to identify the source of themobile phone image. According to the characteristics of mobile phone camera imagingpipeline and analyzes the inherent defects and noise of the imaging sensor, then extractmobile phone camera fingerprint features. In view of the existing image forensics methodbased on pattern noise source a new algorithm has been proposed, the main researchcontributions are listed as follows.Firstly analysis of the similarities and differences between the mobile phone cameras anddigital cameras, due to the pros and cons of CMOS and CCD sensors largely determine thedifference between the mobile phone cameras and digital cameras. The limit of the quality,volume and cost in cell phone camera has resulted in stronger fingerprint characteristics than adigital camera, though the display of the Fourier transform spectrogram of a pixel line in thereference pattern noise show that it has a more obvious intensity gradients and fluctuant effectthan a digital camera, which do a great extent to the blind forensics of digital image.Secondly, the correlation probability distribution model and the extraction method ofpattern noise to be optimized, due to a large part of the current cameras have appeared theasymmetrical distribution, use the optimized asymmetrical distribution function to match thecorrelation distribution histogram and reduced the minimum error rate effectively. To extractthe pattern noise we proposed a method which according to the optical responsecharacteristics and special influence of the light sensitive points, and use it to make up largecomponents of imaging sensor pattern noise mainly for mobile camera origin identificationaccording to the optical response non-uniformity principle. Experiments show that theproposed method can not only identify the source of cell phone camera images effectively, butalso greatly decrease the computational load for detection.Finally, according to the insufficient of the algorithm based on large component ofpattern noise before, we proposed a method which introduced the CFA interpolation principleto the extraction of pattern noise. Affected by the interpolation points, the pattern noiseextracted before was much larger than the true value. Therefore, we separate the interpolationpoints from the non-interpolated points before the extraction of pattern noise to avoid theirinterference, and then further improve the accuracy of the pattern noise. The experimentalresults show that this method can reduce both false alarm and residual rate in the detectionprocess effectively, and also make two kinds of data boundaries more apparent yet reduce thefalse detection rate, and then improve the detection efficiency.