Reversible watermarking schemes based on expansion embedding using pixel partition strategies

Abstract

Reversible watermarking methods have drawn attention of many researchers in recent years because original cover objects can be completely restored upon extraction of embedded information. The prediction error expansion (PEE) embedding proposed by Thodi and Rodriguez can embed one information bit into one cover pixel and achieve good visual quality of watermarked images. However, the PEE method is not a generalised scheme and its embedding capacity measured by bit per pixel (bpp) and visual quality measured by peak signal-to-noise ratio (PSNR) can be improved. Thus, we propose two novel schemes based on the PEE method by using pixel partition strategies to improve embedding capacity and/or visual quality. The first proposed scheme makes the PEE method be generalised and the second proposed scheme can embed more than 1 bpp. Experimental results show that the proposed schemes can significantly improve visual quality of the watermarked images produced by the PEE method when embedding capacity falls within the range [0·1, 1]. In particular, when an embedding capacity belongs to the range [0·1, 0·5], in average, the first proposed scheme with K = 1 can improve the PSNR of the PEE method at least 13 dB for test images. In addition, when embedding capacity falls within the range [0·1, 1], in average, the second proposed scheme with K = 1 can improve the PSNR of the PEE method at least 14 dB for test images.

Keywords:

• reversible watermarking
• difference expansion
• prediction error expansion
• expansion embedding
• reversible integer transform
• data embedding
• information hiding