A self-embedding technique for tamper detection and localization of medical images for smart-health

摘要

With the prodigious headway of the Internet of Things (IoT), cloud computing, Artificial Intelligence (AI), and big data, smart healthcare is expected to provide potential and competent healthcare services. Smart healthcare is changing the traditional healthcare system by making it more convenient, more expedient, more effective, and more personalized. The rampant health sector data breaches worldwide, however, testify to the need of ensuring the integrity and authenticity of data shared over insecure networks. In this paper, a secure self-embedding fragile watermarking scheme capable of authenticating the medical images and precisely localizing the tampered regions is presented. Two watermarks generated from the cover image called authentication watermark and localization watermark, are used for authentication and localization of the tampered region at the receiver. For watermark generation, the cover image is divided into 4 x 4 non-overlapping blocks. Each block is permuted using chaotic encryption before the watermark generation. The authentication watermark is a function of the 4-Most Significant Bits (MSBs) of each pixel of a block. Deoxyribonucleic Acid (DNA) encoding is used to ensure the security of the authentication watermark before its embedding. The localization watermark utilizes the arithmetic mean of a selected block and the Maximum Pixel Intensity (MPI) in that block. The DNA arithmetic is applied to generate the final authentication of watermark data. The tamper detection and localization results obtained for the proposed work are found to perform better compared to the state-of-art techniques. The proposed algorithm maintains better visual quality despite higher embedding capacity as indicated by an average Peak Signal to Noise (PSNR) value of 51.94 dB for an embedding capacity of 262,144 bits. The average value for the Structural Similarity Index Metric (SSIM) for the proposed scheme is found to be 0.9962 which is higher when compared to the techniques under comparison. The average False Positive Rate (F-PR) for the proposed algorithm is found to be 3.9916 for tampering rates varied from 5 to 50%. The scheme outperforms the various state-of-the-art techniques making it an efficient candidate for tamper detection and localization in smart health applications.