Nonuniform quantizer design for image data hiding

Most quantizer based data hiding schemes use uniform quantizer, which is not optimal if the host signal is not uniformly distributed. In this paper, we design a quantizer that is not only pdf-matched but also more suited to embedding than the Linde-Buzo-Gray (Lloyd-Max) algorithm for vector (scalar) quantizer design. Experimental results shows that the proposed scheme provides better trade-offs between robustness to attacks, embedding induced distortion and embedding capacity than a simple pdf-matched scheme. The proposed algorithm also shows about 3 dB improvements in embedding distortion over other popular quantizer based embedding algorithms.