The Internet is a conduit for vast quantities of digital data, with the transmission of images being especially prevalent due to the widespread use of social media. However, this popularity has led to an increase in security concerns such as image tampering and forgery. As a result, image authentication has become a critical technology that cannot be overlooked. Recently, numerous researchers have focused on developing image authentication techniques using deep learning to combat various image tampering attacks. Nevertheless, image authentication techniques based on deep learning typically classify only specific types of tampering attacks and are unable to accurately detect tampered images or indicate the precise location of tampered areas. The paper introduces a novel image authentication framework that utilizes block-wise encoding through Variational Autoencoder and Generative Adversarial Network models. Additionally, the framework includes a classification mechanism to develop separate authentication models for different images. In the training phase, the image is first divided into blocks of the same size as training data. The goal is to enable the model to judge the authenticity of the image by blocks and to generate blocks similar to the original image blocks. In the verification phase, the input image can detect the authenticity of the image through the trained model, locate the exact position of the image tampering, and reconstruct the image to ensure the ownership.
