Explainable Machine Learning Models and Architectures

7Image Forgery Detection: An Approach with Machine Learning

Madhusmita Mishra¹*, Silvia Tittotto² and Santos Kumar Das¹

¹Dept. of ECE, NIT Rourkela, Rourkela, Odisha, India

²CECS, UFABC, Santo Andre, Sao Paulo, Brazil

Abstract

Before the camera was invented, verification processes were done manually. With the advancement in technology, now it is impossible to predict the difference between the actual and edited image. This is because of the growing varieties of advanced image editing tools. These editing tools have low cost and are open-source tools meant for the handlers and commonly utilized to create memes for uploading to social media websites. The present work illustrates forgery detection of an image via the error level analysis (ELA) method. Here the binary decision of convolutional neural networks (CNN) based model is taken in declaring the aptness of image intended for official applications. In the CNN model, trained Kaggle dataset is applied. The comprehensive simulations confirm the accuracy and precision of the suggested model.

Keywords: CNN, image forgery detection, machine learning (ML), error level analysis (ELA), convolutional neural network (CNN), deep learning (DL), joint photographic experts’ group (JPEG)

7.1 Introduction

Images provide much information to the viewer. In the present-day societal media and the internet, digital images have undoubtedly evidenced their powerfulness and suitability. The need for images is never-ending in medical imaging, intelligence, ...