In this work, the Ph.D. candidate Anselmo Ferreira along with professors Jefersson dos Santos (UFMG) and Anderson Rocha (RECOD) propose multi-scale and multi-directional median filtering detection algorithms to identify the presence of image tampering in digital images.
These algorithms are rooted at the hypothesis that median filtering streaking artifacts disturb the image quality of median-filtered images in a specific way under multi-scale filtering and over progressive perturbations.
The technique is different from others in the literature as it progressively perturbs the image by blurring it multiple times with different kernels, building a discriminative feature for later decision making by using image quality metrics. Experiments with complex datasets show that the proposed method outperforms state-of-the-art solutions and represent an important step toward developing more robust and reliable forgery detection methods.
The pre-print is available here.
FERREIRA, Anselmo; DOS SANTOS, Jefersson Alex; ROCHA, Anderson. “Multi-directional and Multi-scale Perturbation Approaches for Blind Forensic Median Filtering Detection”. Intelligent Data Analysis (Print), 2016.