A Convolutional Neural Network Based Seam Carving Detection Scheme for Uncompressed Digital Images

Revealing the processing history that a given digital image has gone through is an important topic in digital image forensics. Detection of seam carving, a content-aware image scaling algorithm commonly implemented in commercial image-editing software, has been studied by forensic experts in recent years. In this paper, a convolutional neural network (CNN) architecture is proposed for seam carving detection. Unlike the existing forensic works in detecting seam carving, where the feature selection and the pattern classification are two separated procedures, the proposed CNN-based deep learning architecture learns and then uses more effective features via joint optimization of feature extraction and pattern classification. Experimental results conducted on a large dataset have demonstrated that, compared with the current state-of-the-art, the proposed CNN based deep learning scheme can largely boost the classification rates as the seam carving rate is rather low.