Non-overlapping block-level difference-based image forgery detection and localization (NB-localization).

With advent of digital devices, we are surrounded by many digital images. We usually believe on digital images in whatever form presented to us. Therefore, we need to be careful as the images may be forged. There exist several image forgeries through which original intent of the image may be hidden and some other meaning is reflected through forgery. Copy-move forgery is one such forgery technique, where the manipulator copies certain portion of the image and duplicates it in some other portion of the same image. In this paper, we propose a novel approach to detect the copy-move forgery in images using non-overlapping block level pixel comparisons and that can achieve better detection and classification accuracy. This approach divides image into 4, 5, 6 or more such blocks and compare each block by moving sliding window over the entire image which is not overlapping with current block. It was found that with different number of blocks the forged region of different sizes can be easily found. We have used SSIM (structure similarity index) parameter to classify the image as forged or original. Algorithm is simulated on various datasets including (MICC, CASIA, coverage, and COMOFOD, etc.) and achieved maximum accuracy of 98% and also compared our result on precision, recall, FPR and FNR including other parameters. [ABSTRACT FROM AUTHOR]