Image forgery detection using error level analysis and deep learning
Ida Bagus Kresna Sudiatmika, Fathur Rahman, Trisno Trisno, Suyoto Suyoto
Many images are spread in the virtual world of social media. With the many editing software that allows so there is no doubt that many forgery images. By forensic the image using Error Level Analysis to find out the compression ratio between the original image and the fake image, because the original image compression and fake images are different. In addition to knowing whether the image is genuine or fake can analyze the metadata of the image, but the possibility of metadata can be changed. In this case the authors apply Deep Learning to recognize images of manipulations through the dataset of a fake image and original images via Error Level Analysis on each image and supporting parameters for error rate analysis. The result of our experiment is that we get the best accuracy of training 92.2% and 88.46% validation by going through 100 epoch.