Deep Learning based Zero Watermarking for Authentication of Medical Records

Abstract

The security of digital images is crucial since they often contain sensitive and confidential data. Unauthorized access to this data could result in severe penalties for the parties involved. Despite the availability of highly secure algorithms, security remains a significant concern due to the rapid emergence of new technologies that can breach it. Thus the proposed work implements a technique that makes the confidential data inaccessible to intruders. Hence fragile type of data hiding technique is used where even with the slightest tampering to the image by an attacker, the information i.e. watermark image is completely destroyed, hence preventing it from unauthorized access. Also, a hybrid transform including DTCWT and NSST is used to fuse two medical images to form a more sophisticated output image, which serves as the final watermark. Further, the zero watermarking model is implemented using the ResNet 50 DL model for more precise results and extraction of feature maps. Embedding the actual image in the carrier image could make the watermarking detectable especially when it is fragile, hence Zero Watermarking overcomes this also by virtual embedding. Moreover, the algorithm employs the avalanche effect of SHA512 for highly secure authentication, further strengthening the security of the system. Overall, the proposed method is an effective way to ensure the security of digital images with confidential data.