Image Splicing Detection Using Support Vector Machine (SVM)

Abstract

The official establishment's Fake images pose a real problem across the internet and mostly involve social media to represent cybercrime and misinformation. At present, a lot of reactions are based on the images.  The objective of this research is to evaluate the accuracy of image splicing detection based on the true positive and false positive rates by implementing techniques using machine learning and leveraging features such as HOG, ELA, DWT, and LBP and classified with SVM. This research used images as their datasets within the two benchmark datasets CASIA v1.0 and CASIA v2.0 and it highlights the potential of integrating the feature extraction techniques for robustness, accuracy, and reliability. However, this research found that result that with combination feature extraction ELA and HOG achieved the highest accuracy with 94.00% for CASIA v1.0 followed by CASIA v2.0 at 95.50%, showing that they performed well balancing accuracy. Moreover, from the side of HOG and LBP, it showed the lowest performance accuracy for imbalance detection, and LBP and DWT yielded moderate accuracy. Overall, these findings demonstrate the ELA and HOG combination feature extraction had reliable outcomes for image splicing detection.