banctec eduardo


An evolutive approach for the quick bio-watermarking of digital documents

Eduardo Vellasques, Ph.D candidate

Outline of the proposed research

The fundamental research problem that this project aims to address is the enforcement of (1) integrity, (2) authenticity and (3) confidentiality of document images in a large scale scenario. The management of massive volumes of digital documents (images) is a key part of our digital economy. For example, Banctec customers across the globe scan and store over 50 million documents (bank cheques, invoices, etc) in a single day. Protecting these documents is a challenging task. The classical computer security methods (e.g. cryptography) cannot fully protect such type of data because they work like a barrier (the protection vanishes when the barrier is surpassed). Digital watermarking, which consists of embedding information in an image through changes on its pixel values, allows tackling issues (1) and (2) since the protection is fused with the image (issue (3) can be easily tackled with the use of shuffling). However, it has been observed in the literature that the embedding involves two conflicting objectives: watermark robustness (against attacks) and quality of the watermarked images. This is clearly an optimization problem. Many authors have employed evolutionary optimization algorithms to address this problem, in a static manner (optimizing each image individually). This approach has a prohibitive computing cost.

The long term objective of the proposed project is to develop an adaptive digital watermarking method to protect massive quantities of documents of the same nature. The short term objectives are:

      - Creation of a dynamic multi-objective Particle Swarm Optimization (PSO) algorithm;

      - Creation of a dynamic intelligent watermarking algorithm, which uses a long-term memory to boost the optimization speed.

The research hypothesis is that while watermarking massive quantities of documents of same nature, the optimum watermarking parameters of a given image is somewhat similar to the optimum watermarking parameters of the previous images. To accomplish the proposed objectives, this hypothesis will be firstly validated experimentally. After that, the existing dynamic and multi-objective (MO) evolutionary techniques will be evaluated individually and then, combined into a whole new Dynamic MOPSO method. This method will be evaluated experimentally with the use of the baseline watermarking system and the large scale database (provided by Banctec and comprising over 100K samples of real banking documents). Finally, a mechanism will be created to handle the dynamic part of the proposed method and evaluated in the proposed database.

This project will contribute to the advancement of knowledge with the inception of a dynamic multi-objective/multi-criteria optimization method applied to digital watermarking. Among other contributions I will formulate the problem of intelligent watermarking as a dynamic optimization problem. This knowledge will make possible the industrial use of digital watermarking, which is considered strategic by the industry. Banctec has demonstrated interest in patenting such methods, and have contributed with consulting and access to digital scanning equipment. This project is also financed by NSERQ-CRSNG.


Vellasques, E., Sabourin, R. and Granger, E., Intelligent Watermarking of Document Images as a Dynamic Optimization Problem, The Sixth International Conference on Intelligent Information Hiding and Multimedia Signal Processing (IIH-MSP 2010), Darmstadt, Germany, October 15-17, 2010.


[1] Hassanien, A. and al. (2008) Computational Intelligence in Multimedia Processing: Foundation and Trends, in Computational Intelligence in Multimedia Processing: Recent Advances, Series: Studies in Computational Intelligence, Vol. 96, Springer.
[2] Muharemagic, E. (2004) Adaptive Two-level Watermarking for Binary Document Images, PhD thesis, Florida Atlantic University.
[3] Shieh, C.S. and al. (2004) Genetic Watermarking Based on Transform-domain Techniques, Pattern Recognition 37, pp 555-565.