Halftone Image Watermarking Using Iterative DBS Method
|Event Date:||August 21, 2014|
|Speaker:||Professor Fuping Wang|
|Speaker Affiliation:||Department of Electrical Engineering, Tsinghua University, Beijing|
|Contact Name:||Professor Jan Allebach
Printed images include passport photos, coupons, and tickets. There are a number of applications in which halftone image watermarking techniques can play an important role for printed images, e.g. copyright labeling, forgery prevention, and documenting ownership.
A number of watermarking methods for halftone images have been developed using screening or error diffusion halftoning techniques. Although these methods can often be effective, they are under pressure to improve watermarked halftone image quality when the watermarking capacity is high. The strategy of halftone quality control in these methods, to my point of view, are in open-loop form, and are therefore not able to ensure optimal performance.
The direct binary search (DBS) algorithm, which employs a search heuristic to minimize the mean-squared perceptually filtered error between the halftone and continuous-tone original images is more computationally intensive than screening or error diffusion, but yields significantly better output quality. In many applications of printed image watermarking, the computational burden of embedding is not as serious a concern, as long as the watermark decoding step can be efficiently implemented. On these occasions, DBS is preferred for halftone watermarking; and superior performance can be expected.
I will present my research work on halftone image watermarking using the DBS method during my half-year visit in ECE at Purdue University. In my research, watermarking and halftoning are two optimization goals. And the way to embed the watermark and the method to solve the optimization problem are two critical issues. Two watermark embedding approaches: cell-pair correlation embedding and pixel-pair correlation embedding, are proposed. And two optimization methods: weighted sum cost function optimization and optimization with a strict constraint, are formulated and solved, accordingly. Simulation results with different levels of embedding capacity show that: (1) the quality of the watermarked halftone is comparable with that of the unwatermarked image when the embedding capacity is about 10%; and (2) more surprisingly, the watermarked halftone is more visual appealing than the unwatermarked halftone when the embedding capacity is about 5%. This is likely due to escaping local minima during the search when watermarking is included.
As a farewell report, I would like to discuss some less formal and not purely academic matters. I would like to recall some details when I was doing the research directed by my host: Professor Jan P. Allebach.
Fuping Wang received the B.S. and Ph.D. degrees in Electrical Engineering in 1997 and 2002, respectively, all from Tsinghua University. He was a Postdoctoral Fellow in the Department of Automation at Tsinghua University from April 2002 to April 2004, and a Lecturer in the Department of Electrical Engineering at Tsinghua University from May 2004 to December 2007. Since January 2008, he has held the rank of Associate Professor in the Department of Electrical Engineering at Tsinghua University. Since January 2014, he has been a Visiting Scholar at Purdue University. Professor Wang’s research interests include communications, chaos-based data signal processing, encryption, chaos time-series analysis, and watermarking.