Robust median filtering forensics using texture feature and deep fully connected network

Image forensics researchers have recently focused a lot of attention on detection of median filtering that is used to hide evidence of image forgery and related operations. After median filtering, JPEG compression, Gaussian blur, noise addition, and resampling are often applied to further hide traces of median filtering. Researchers are essentially abandoning a well-developed and established field of feature engineering for deep learning-based system that do not require manual feature engineering. In this paper, we combined the feature engineering and machine learning to develop a solution for robust median filtering detection. For the purpose, we developed an 18-D feature based on textured features and a neural network called MedFCN to performing classification between original and post-median filtering modifications by operations such as Gaussian blur, resampling, the addition of additive white Gaussian noise (AWGN), and JPEG compression. All datasets used in the study were created from UCID, BOSS, RAISE, and Dresden for training and testing of the system. By designing a multilayer perceptron that can accept an engineered feature set, we were able to combine two methodologies, feature engineering and deep learning, to develop a system that outperforms state-of-the-art methods for median filter forensics. © 2023, The Author(s), under exclusive licence to Bharati Vidyapeeth's Institute of Computer Applications and Management.