Regularized CNN Model for Image Forgery Detection

Abstract

Digital images play a very important role in different areas in the modern technological scenario. Changing and manipulating the content of the digital image is a very easy task by using powerful image editing tools. In today's technology environment, digital photographs serve a critical function in a variety of fields. Using advanced image editing tools, changing and rearranging the content of a digital image is a simple operation. It is now possible to add, edit, or remove essential aspects from an image despite leaving any perceptible alterations. In addition to determining if the picture is authentic or forged, the metadata of the image may be examined, however, metadata can be altered. In this example, the authors use Error Level Analysis on each picture and matching parameters for error rate analysis to detect images of modifications using Deep Learning on a dataset of a false image and real photos. This experiment shows that by running through 100 epochs, we obtain the best training accuracy of 99.17 % and 95.11 % of accuracy validating.