Data hiding is an important way of realising copyright protection for multimedia. In this study, a new predictive method is proposed to enhance the histogram-based reversible data hiding approach on grey images. In those developed histogram-based reversible data hiding approaches, their drawbacks are the number of predictive values less to the number of pixels in an image. In these interleaving prediction methods, the predictive values are as many as the pixel values. All predictive error values are transformed into histogram to create higher peak values and to improve the embedding capacity. Moreover, for each pixel, its difference value between the original image and the stego-image remains within +1. This guarantees that the peak signal-to-noise ratio (PSNR) of the stego-image is above 48 dB. Experimental results show that the histogrambased reversible data hiding approach can raise a larger capacity and still remain a good image quality, compared to other histogram-based approaches.

This paper proposes a hierarchical fragile watermarking scheme for image authentication with localization and recovery. Two phases are exploited for pixel-wise and block-wise authenticity. By applying the singular value decomposition (SVD), only a few pixels need to be modified to carry these watermark bits so we can produce high quality of the watermarked image and achieve the integrity verification of blocks. Pixel-wise tampering detection and recovery is also realized in this proposed scheme. Altered pixels in each block can be exactly localized using the authentication bits. Accordingly, unaltered pixels remain unchanged but only altered pixels need to be recovered. In the image recovery stage, instead of replacing the whole block identified as tampered, elements belonging to associated VQ codeword are restored for those altered pixels. In this way, the quality of recovered image is better than replacing whole altered block by a whole vector quantization block. The experimental results reveal that the average PSNR values of reconstructed images are higher than that of other schemes.

No abstract