Machine Learning Based Image Forgery Detection using Feature Fusion of Otsu Binarization and Thepade SBTC

Abstract

Digital images are essential in every field, including clinical imaging, media broadcasting, crime analysis, and scientific research. The development of robust image editing software has simplified the process of manipulating photographs and changing their content. The image may now have important aspects added, modified, or removed without leaving any visible indications of manipulation. As a result, there is a need to design reliable methods to detect such manipulations. The paper proposes a technique for detecting tampered images using machine learning models trained on feature vectors generated by a fusion of the local features generated with the Otsu binarization technique and global features formed with Thepade Sorted Block Truncation Coding (Thepade SBTC). The proposed forgery detection methodology is empirically validated on the MICC-F220 dataset of 220 photos (with equal tampered and genuine images) using ten machine-learning classifiers and four ensembles. The best performance is given by the majority voting ensemble of Random Forest+ Random Tree + IBK with the feature fusion of Otsu binarization with Thepade SBTC 10-ary features. The fusion of features has shown better image forgery detection capability over consideration of individual features.