Abstract: Because of the difficulty of making adequate redundant capacity for watermarking, some reversible watermarking schemes for content authentication just embed a symmetrically encrypted hash value, the size of which is significantly smaller than that of a digital signature. It makes a dishonest verifier be capable of fabricating legal contents. However, the research finds that introducing a public-key signature, which is usually several times longer than a hash code, can only need several bytes more capacity if the subliminal channel in the signature scheme can be used. And it is shown that RSA-PSS (RSA-probabilistic signature scheme) has such an exploitable subliminal channel. Then an improved reversible R-S (regular-singular) watermarking scheme, which adopts RSA-PSS, is proposed. It still produces the redundant watermarking capacity by compressing the R-S vector but stores partial or entire compressed R-S vector in the subliminal channel. As a result, the new scheme only needs 4 bytes more additional capacity. By the reduced size of overall embedded data, the new scheme can even partition an original image into several tens of blocks and embed data block-wise so that it has the capability of localizing tampering. And by not revealing any information directly about the private key, it has reliable security also.