A Robust Algorithm for Information Hiding in Digital Pictures

A problem inherent in information hiding techniques for digital images is that of alignment during data extraction. This alignment problem can be solved using a search approach where a combination of coarse orientation detection, random search and gradient descent methods are employed. When used in conjunction with the Patch Track algorithm, it is possible to create an information hiding and retrieval system that is robust towards rotation, cropping and noise; successful data extraction can occur with a high degree of certainty from scanned images, rotated images and partial images. An image can be encoded with a desired piece of information (a reference number, URL etc.) with Patch Track by way of pseudo-random, imperceivable alterations of intensity throughout the image. This information can be then be extracted by the intended parties with a decoder. Various applications exist for this watermarking system including photo annotation, verification of image ownership on the Internet and data warehousing. Thesis Supervisor: Walter Bender Title: Assistant Director of Information Technology, Senior Research Scientist, The MIT Media Laboratory