Enhancing the Image Forgery Detection based Machine Learning Approach using Multiple Datasets

Abstract

Nowadays, detecting forged images has become increasingly important because of the widespread use of advanced image editing tools. Splicing is one common form of forgery, where parts or different images are combined to create misleading images. However, detecting this type of forgery poses a challenge because it often appears highly realistic and is difficult to distinguish from authentic images. This study presents a method for detecting forged images. The proposed system aims to enhance forgery detection by carefully analyzing images using preprocessing, such as resizing, converting colors to HSV, analyzing histograms, converting images into binary numeric values, and visualizing the original and forged images and their respective hues based on grayscale, RGB, and HSV histograms. The proposed method used three machine learning algorithms, namely Multilayer Perceptron (MLP), Support Vector Machine (SVM), and K-Nearest Neighbors (KNN), and the results demonstrate its effectiveness in rapidly discerning forged images while maintaining high accuracy of 99.72% on the MISD, 99.53 % on the CASIA2, 97.44 % on the NC2016, and 94.30 % on the CoMoFoD datasets.