THREE PHASE DIGITAL IMAGE REVERSIBLE WATERMARKING WITH LOCALIZED TAMPER DETECTION

Abstract

Reversible watermarking is a lossless watermarking technique, which allows the extraction of watermark as well as the recovery of original cover media at the receiver end for ensuring authentication and integrity verification of digital data. In this project, a high capacity improved median based three phase reversible watermarking algorithm using prediction error expansion for grayscale digital images is proposed. The scheme predicts the candidate pixels using median of original neighboring pixels, which keeps magnitude of prediction errors relatively small, also the prediction errors in particular phase are sorted according to context pixels variance to obtain low distortion for small payloads. Furthermore, I have shown the efficient creation of location map that requires only single bit to test for overflow/underflow. The performance of the scheme is analyzed for some of the publicly available standard and medical test images in terms of maximum embedding capability, PSNR, SSIM index values and location map size. The scheme is also compared with some of the state-of-the-art schemes. The comparison results demonstrate that the proposed scheme has lesser distortion for different payload. Further, in this project I have also presented localized tamper detection in the received digital image as one of the application of the proposed scheme which allows the receiver to reject only the selective regions in case of integrity failure. In this application, the image is divided into blocks, the integrity value is then computed taking into account all the block pixels and the corresponding block number, subsequently this value is reversibly embedded in the block. The current and next blocks are combined dynamically when embedding capacity of a block is not sufficient. The experimental results on some test images indicate that the watermarked image has high PSNR and SSIM value also the results under various attacks demonstrates the effectiveness of the algorithm.