Median filtering detection using optimal multi-direction threshold on higher-order difference pixels

Image forgery detection is a challenging issue because fake images can be prepared accurately using precise editing tools. In this paper, a robust image forensics technique based on a multi-direction threshold (MDT) is proposed to detect median filtering. In the proposed method, an optimal thresholded array is derived from difference arrays in multiple directions. The Markov process is applied to fetch the joint probability statistics of neighboring pixels on optimally thresholded difference arrays. The proposed optimal MDT utilizes both first and second-order difference arrays for additional information that leads to a more comprehensive feature set. As a result, the proposed technique achieves 93.40%, 90.59%, and 85.76% detection accuracy on JPEG compressed images of size 64 x 64 pixels with quality factors 70, 50, and 30, correspondingly. The non-filtered and median filtered images are classified using LDA and SVM classifiers. The superiority of the proposed technique is analyzed through exhaustive experimental analysis on centrally cropped and zero-padded median filtered images.