A Noise Convolution Network for Tampering Detection

The vulnerability of digital images to tampering is an ongoing information security issue in the multimedia field. Thus, identifying tampered digital images and locating the tampered regions in the images can help improve the security of information dissemination. A deep fusion neural network named NC-Net is designed in this paper, introducing pattern noise as assistance to fully exploit the tampered features present on the tampered image. The incorporation of noise texture information enabled NC-Net to acquire deeper tampered image features during the training phase. The extracted noise is incorporated as a crucial component within the convolutional structure of the model, serving as a potent activation signal for the tampered region. The performance of NC-Net is confirmed through relevant experiments on publicly available tampered datasets, and outstanding results are achieved in comparison to other methods.