Image Manipulation Detection and Localization

Abstract

Image manipulation along with image splicing imposes a significant problem in the real world as it involves the act of maliciously combining parts from different images to create a forged image. This can lead to the spread of misinformation, deceptive visual narratives, and the decline of trust in authentic visual media. To address this challenge, a solution has been implemented by us which explores and evaluates various pre-processing methods such as ELA, JPEG compression, for the manipulation detection part and use of RGB, DCT channels along with CNN for the manipulation localization with the help of CASIA dataset. Our emphasis lies in the innovative application and integration of these methodologies to achieve a seamless and effective solution. The system takes in data in the form of an image file (JPG, JPEG, PNG), resizes it and feeds it to the machine learning algorithm. Experimental results are demonstrated that include training accuracy, validation (test) accuracy and loss.