“…The features are then used to train a SVM classifier. The results on benchmark databases showed that this scheme is capable of detecting JPEG measures grainy noise due to anti-forensic dither in DCT domain using TV, accuracy almost 100% over QF ∈ [30,84] Wang et al [126] 2014 quantisation UCID [78](1338 greyscale 512×384 images, compression rate from 0.5 to 8bpp, SVM, JPEG2000, SPIHT)…”