Image Forgery Detection using Machine Learning with Fusion of Global and Local Thepade's SBTC Features

Abstract

Image forgery is manipulating digital images to hide or change some useful information contained in the images. Images are considered the most effective way to convey information, and manipulating this information sometimes creates havoc. The action of tampering with images that are done either for fun or to give false evidence has resulted in a disaster in some cases. It is done in such a way that it cannot be determined by the naked human eye, so many people have implemented various types of machine learning algorithms, which they have implemented with handcrafted features to determine different types of forgery and whether an image is forged or not. These algorithms are used to extract the digital signature differentiating whether an image has been tampered with or not. Various techniques have been implemented for either fine or coarse image splicing, whereas a technique dealing with both needs to be devised. For this, our proposed work focuses on different types of machine learning classifiers and 10-fold classification. The attempted values of n for the machine learning classifier include 2,3,4. The different types of classifiers include Random Forest, Random tree, support vector machine, Logistic Regression, Naive Bayes. These classifier models are trained on comofod, casia v2.0 datasets. Accuracy increase is observed when a fusion of Thepade's Sorted Block Truncation Coding (i.e., Thepade's SBTC) local and Thepade's SBTC global feature tables.