Towards Lower Bounds on Distortion in Information Hiding

Abstract

Information hiding refers to embedding additional digital data in cover objects–e.g. audio, image or video signals–by modifying the cover objects. Information hiding techniques have been used in a variety of application domains including copyright protection for digital media, content authentication, media forensics, and covert communications. The goal of information hiding techniques is to minimize the effect that the embedding process has on the cover object. Embedding processes introduce distortion to cover images and change the appearance as well as statistics of images. This is undesirable in many applications including fragile watermarking and steganography. In this dissertation, information hiding algorithms using JPEG images have been proposed with three goals: minimizing distortion due to embedding, preserving statistical properties during embedding, and predicting a distortion level for a given message length. Proposed methods use rounding errors created at the JPEG quantization step as side information, and the methods are based on block-based coding techniques known as parity coding and matrix coding. A mathematical analysis predicts the distortion introduced by each proposed algorithm as a function of message length and the rounding error distribution of the cover image. Minimization of distortion in proposed methods was experimentally validated. Extensive tests using state-of-the-art steganalysis software show that the proposed information hiding methods compare favorably to other published methods for given embedding rates.