Median Filtering Detection Based On Local Binary Patterns

Since digital multimedia, such as digital pictures and digital videos, behaves the property of easy manipulation, copy and spread, the amateur can effectively edit or tamper with digital multimedia resources by means of common multimedia processing and editing software, without leaving visual clues of forgery. In this environment, the passive blind image forensics has attracted wide interest in the field of information security techniques.Based on the digital image as the research object, this dissertation focuses on a common form in digital image processing and manipulation----median filtering. Currently, the main problem for this research is the large dimension of the extracted characteristics and high algorithm complexity. In this paper, we propose a novel algorithm for detecting median filtering in digital images based on hierarchical local binary pattern(LBP) features, which can improve the above problems effectively when maintain a high accuracy.The main contents include:(1) The detection of median filtering processing and its window size: starting from the characteristics of median filtering, this paper analyzes the characteristics of different local feature descriptor, then chose local binary pattern which is sensitive to local variation as the base of median filtering detection. After improved it, two kinds of enhanced LBP operators(E-LBP and N-LBP) are defined according to the relationship between adjacent pixels in the local(9-dimension). This algorithm not only can realize the detection of image median filtering, but also can detect the median filter order.(2) The recognition of median filtering shape: in view of the several common median filter shapes(square, cross and fork), this dissertation puts forward a set of 5-dimension features on the bases of E-LBP, it can further determine its filtering window shape when the image is identified as median filtering.The innovation of this article is combining the local texture operator LBP with digital image tampering detection, and can effectively detect whether an image had processed by median filtering or not; at the same time, this paper proposed a 5-dimension features to probe the shape of the median filter. The experiments show that, compared with other traditional algorithms, the proposed algorithm can not only reduce the computation time, but also can acquire a higher detection rate with low dimension. It can be used to explore the order and window shape of the median filters effectively and achieve very good effect of median filtering detection.