Audio Reversible Watermarking Scheme in the intDCT Domain with Modified Prediction Error Expansion

Reversible watermarking schemes (RWS) allow the restoration of the original signals after the watermarks are extracted. Most RWS for audio signals use time-domain for information hiding, although their transparency is hard to maintain for high embedding capacities. Some audio RWS use the frequency domain to improve transparency; however, their embedding capacity is lower than that of time-domain schemes. In this manuscript a RWS for audio signals is proposed, it differs from other schemes that work with the intDCT domain in the use of auditory masking properties, which are exploited to improve transparency, and the increase on embedding capacity is explored through a modified prediction error expansion (PEE). The payload capacity is 27.5 kbps with a degradation over -2 ODG, which are adequate results for practical audio applications. A generalized multi-bit expansion is proposed and experimental results suggest that higher expansion factors improve transparency.