Protecting Tampered Images with Lossless Auto Recovery using improved INN

The Image Immunizer project presents a novel strategy to address the risk of manipulation associated with digital photos. It uses Vaccinator and Invertible Neural Network technologies to provide a complete solution that guarantees tamper resistance and lossless auto-recovery. Four essential components make up the multitask learning network: self-recovery encouragement, consistency checking, tampered pixel identification, and image vaccination. An immune version is produced by applying changes to the original image and its edge map during the forward pass. A localizer uses a predicted tamper mask to identify an object precisely when tampering is present. Using Run-Length Encoding, the backward pass preserves image integrity by transforming concealed disturbances into recoverable information. Extensive real-world testing demonstrates the efficacy of the system in accurately correctly localizing and recovering high-fidelity content, skillfully handling the difficulties presented by low-resolution or compressed images. The Image Immunizer stands as a robust solution against the diverse tampering techniques encountered in practical scenarios.