Progressive quantized projection approach to data hiding.

Abstract

A new image data-hiding technique is proposed. The proposed approach modifies blocks of the image after projecting them onto certain directions. By quantizing the projected blocks to even and odd values, one can represent the hidden information properly. The proposed algorithm performs the modification progressively to ensure successful data extraction without the need for the original image at the receiver side. Two techniques are also presented for correcting scaling and rotation attacks. The first approach is an exhaustive search in nature, which is based on a training sequence that is inserted as part of the hidden information. The second approach uses wavelet maxima as image semantics for rotation and scaling estimation. Both algorithms have proved to be effective in correcting rotation and scaling distortion.