Median filtering detection using LBP encoding pattern

In recent years, median filtering detection has a widely application in many fields such as images' processing history tracking, image editing detection, image anti-forensics analyzing and anti-steganalysis analyzing. In this paper, we propose two median filtering detection algorithms. The Algorithm I is a recognition algorithm that can identify whether a given image has undergone median filtering. The Algorithm II is a discriminating algorithm that can distinguish a median (average, Gaussian) filtered image from unfiltered images. Differing from the general framework of existing median filtering detectors, the contribution of our work is that the presented methods are not based on the statistical learning model. The proposed methods do not need any classifier, or any threshold. These methods are implemented by counting the number of specific Local Binary Pattern encoding patterns of a single image. Experimental results demonstrate that the proposed methods provide high accuracy and broad-spectrum robustness for tolerating content-preserving manipulations. Compared to state-of-the-art methods, the proposed methods exhibit high efficiency, high accuracy, and strong robustness.