Reversible data hiding with context modeling, generalized expansion and boundary map

Abstract

This paper proposes a reversible data hiding scheme with high capacity-distortion efficiency, which embeds data by expanding prediction-errors. Instead of using the MED predictor as did in other schemes, a predictor with context modeling, which refines prediction-errors through an error feedback mechanism, is adopted to work out prediction-errors. The context modeling can significantly sharpen the distribution of prediction-errors, and benefit the embedding capacity and the image quality. To expand prediction-errors, the proposed scheme utilizes a generalized expansion, which enables it to provide capacities larger than 1 bpp (bits per pixel) without resorting to multiple embedding. Besides, a novel boundary map is proposed to record overflow-potential pixels. The boundary map is much shorter compared with either a location map or an overflow map even though it is not compressed. The combination of the context modeling, the generalized expansion and the boundary map makes the overall scheme efficient in pursuing large embedding capacity and high image quality. Experimental results demonstrate that the proposed scheme provides competitive capacity compared with other state-of-the-art schemes when the image quality is kept at the same level.