Digital Watermarking Using Machine Learning Approaches

Abstract

In recent years digital watermarking has gained substantial attraction by the research community. It promises the solution to many problems such as content piracy, illicit manipulation of medical/legal documents, content security and so on. Watermarked content is usually vulnerable to a series of attacks in real world scenario. These attacks may be legitimate, such as common signal processing operations, or illegitimate, such as a malicious attempt by an attacker to remove the watermark. A low strength watermark usually possesses high imperceptibility but weak robustness and vice versa. On the other hand, different set of attacks are associated with distinctive watermarking applications, which pose different requirements on a watermarking scheme. Therefore, intelligent approaches are needed to adaptively and judiciously structure the watermark in view of the current application. In addition, traditional watermarking techniques cause irreversible degradation of an image. Although the degradation is perceptually insignificant, it may not be admissible in applications like medical, legal, and military imagery. For applications such as these, it is desirable to extract the embedded information, as well as recover the sensitive host image. This leads us to the use of reversible watermarking. An efficient reversible watermarking scheme should be able to embed more information with less perceptual distortion, and equally, be able to restore the original cover content. Therefore, for reversible watermarking, capacity and imperceptibility are two important properties. However, if one increases the other decreases and vice versa. Hence, one needs to make an optimum choice between these two properties for reversible watermarking. 5The research in this work is two-fold. Firstly, we develop intelligent systems for making optimum robustness versus imperceptibility tradeoffs. The performance of the existing watermarking approaches is not up to the task when we consider watermark structuring in view of a sequence of attacks, which is much desirous in real world applications. In order to resist a series of attacks, we employ intelligent selection of both the frequency band as well as strength of alteration for watermark embedding using Genetic Programming. To further enhance the robustness of the watermarking system, Support Vector Machines and Artificial Neural Networks are applied to adaptively modify the decoding strategy in view of the anticipated sequence of attacks at the watermark extraction phase. Secondly, we devise an intelligent system capable of making optimum/ near optimum tradeoff between watermark payload and imperceptibility. In the context of reversible watermarking, we propose an intelligent scheme which selects suitable coefficients in different wavelet sub-bands and yields superior capacity versus imperceptibility tradeoff. Experimental results show that machine learning approaches are very promising in state of the art watermarking applications.