Median filtering detection based on multiple residuals in spatial and frequency domains

Median filtering detection based on multiple residuals in spatial and frequency domains

연구 분야: Cryptography

학회: Signal, Image and Video Processing

As an important research direction in the field of image forensics, median filtering detection has received extensive attention in recent years. Although many methods have been proposed, most of them show insufficient performance when dealing with low-resolution and high-compression images. To address this limitation, we propose a novel median filtering detection method based on multiple residuals in spatial and frequency domains. Specifically, the proposed sharpening residual and sharpen-median filter residual together with median filter residual form a residual set to capture the changes in image content before and after the median filtering. Then, the autoregressive model and Markov model are adopted to extract the detection features from the multiple residuals in the spatial domain. Moreover, to measure the difference of image details between unaltered and median filtered images, the statistical features of the multiple residuals are extracted in the frequency domain. Finally, the extracted features from the spatial and frequency domains are concatenated into a feature set for detection. Extensive experiments demonstrate that the proposed method outperforms the state-of-the-art methods, especially on small-sized and low-quality compressed images.