Hybrid Dual-Channel Input Image Tampering Detection for Scientific Papers

With the development of science and technology, the application of digital image processing and artificial intelligence technology is becoming more and more widespread, which makes it easier and more difficult to detect forged academic images, and academic images have become a high incidence of scientific and technological journal paper forgery. To ensure the authenticity of academic images and scientific reproducibility of experiments, we developed a network called MultiScopeNet. It combines RGB and DCT streams, performs multi-resolution fusion at each layer, and can comprehensively analyze spatial and frequency domain features of images. We also created a new dataset based on an existing tamper detection dataset by applying an image complementation model to train and validate our model against academic misconduct of tampering using generatively forged images. MultiScopeNet significantly outperforms existing state-of-the-art models in dealing with image tampering at different resolutions and sizes.