Digital Image Forensics - example.pptVIP

Digital Image Forensics By:- - Abhijit Sarang - Pankaj Jindal Which of them are digitally manipulated? How can we know? We call a digital image manipulated if either it has been retouched by a photo editing software or has been produced by the software itself. To prevent the former, the owner of the original image may introduce a watermark or a digital signature. But this process may not be feasible every time. Most approaches for detecting digital image manipulation are blind approaches. Our Methodology In [1], the authors argue that the statistical artifacts associated with images generated from cameras is inherently different form that associated with images manipulated by a software. These properties can be captured by analyzing the noise present in the image. Further, a discrete wavelet transform of the image can also be used to obtain some other statistical features . Building the feature vectors Image De-noising Image was filtered using a wiener adaptive filter and a median filter. Neighborhood model of Wavelet sub bands To capture the strong correlation that exists between the wavelet subband coefficient, we find the residual error by building a neighborhood prediction model. Discrete wavelet transform We find the distance of the sub-bands distribution from the corresponding Gaussian distribution. Results Image denoising True Positive = 37/47 False Positive = 21/53 Neighborhood model of Wavelet sub bands True Positive = 32/47 False Positive = 11/53 Discrete wavelet transform True Positive = 34/47 False positive = 15/53 Detecting Fake Regions Detecting abnormal noise patterns in Image Detecting Duplicated Image Regions References Digital image Forensics For Identifying Computer Generated And Digital Camera Images Sintayehu Dehnie, Taha Sencar and Nasir Memon Exposing Digital Forgeries by Detecting Duplicated Image Regions Alin C Popescu and Hany Farid Noise Features for Image Tampering Detection and Steganalysis H