Biowatermarking of grayscale images


Safa Tliba, Ph.D candidate

Introduction

Digital documents are becoming more and more important in our lives. Unfortunately, new threats have emerged to facilitate their consultation, alteration, and illegal redistribution. Hence the growing interest in developing means of safety features to ensure the integrity and authenticity of these documents. Watermarking aim to address this problem by allowing the insertion of verification or identification information in the document to be protected. Traditionally, such information takes the form of a logo or name of the owner of the document, which has a low reliability in comparison with more recent techniques based on biometric features that identify the owner of content in a unique and reliable way. With the new security challenges, the use of biometrics take more and more importance in various areas where the interest to the use this technique in watermarking.

Problem statement

The development of a reliable system for securing digital documents to ensure the authenticity of documents represents one of the biggest challenges in the field of watermarking. The use of biometric features is seen as a potential solution to ensure greater reliability for the verification of content ownership. However, such an application presents many challenges. Indeed, as any watermarking system the compromise between robustness and imperceptibility must be verified. This compromise is accentuated by the fact that the signature should undergo minimal distortion during the transfer and manipulation of the image in order to ensure authenticity verification after extraction of the watermark. To increase the robustness of a watermarking system, we have to increase the size of the information to embed and embed this information in the most significant elements of the image. This leads to great distortion of the host image, which may decrease its value.

On the other hand, the design of signature verification system must take into account the intraclass variability as any system of signature verification, but also, it must be insensitive to distortions caused by the attacks on the watermarking system. Indeed, the choice of features describing the handwritten signature allows good discrimination between signatures from different people, and between authentic and fraudulent signatures for the same person. But on the other hand, features selected should allow the verification system to remain insensitive to distortions due to attacks on the watermarking system.

In the present project, we investigate the use of handwritten signature as watermarking information, to restrict access to a document for people with the appropriate access rights. As can be checked for safe and reliable signature despite the constraints encountered in the field, such as lack of learning data for classifiers, the need for the construction of a fraudulent signature-based algorithms to test signature verification system, and the variability of the signature that makes the choice of discriminant features for learning classifiers very delicate. We must therefore face the challenges of developing a watermarking system and those of the signature verification.

Methodology

In our study we are particularly interested in developing a gray scale image authentication system by handwritten signature. The system will link to a face image of an individual its handwritten signature, which allows the authentication verification of the image, but also allow a bimodal authentication of a person in secure applications, such authentication of civil documents like driver's license, passport or other document containing a face image as a mean of authentication.

To develop this system we work with development platforms already established within the Laboratoire d'imagerie, de vision et d'intelligence artificielle (LIVIA), dedicated to digital watermarking, and signature verification, that we use for experimentation and validation of developed algorithms.

Our approach will include two major parts. The first is to develop an algorithm for robust and reliable watermarking to allow recovery of the watermark with the least distortion possible while ensuring the invisibility of watermarking by considering the following steps:

      - Increase the robustness of the watermarking algorithm in order to allow recovery of the watermark with least distortion possible, especially after applying a lossy JPEG compression. Which is a very common operation and widely used for image transmitting and storing. The medium and low pass filters, being very used in image processing, are also considered. This is done by using quantification for the coefficients of discrete cosine transform, used for watermark embedding.

      - Define the embedding areas of the watermark. These areas should help increasing the imperceptibility of watermarking while preserving robustness. A study on blocks texturing will lead us to rank these blocks from the most receptive to watermarking to least receptive. Then select a number of blocks equal to size of the watermark.

      - Once these areas identified, we perform a frequency selection for watermark embedding that leads to best compromise in terms of robustness and invisibility using genetic algorithm optimization.


Signature verification system is the second part of our project that helps us in one side to generate the watermark from handwritten signature images and another side use of the extracted watermark from host image to verify the authenticity of this image. This is done by considering the following steps:

      - Exploiting the off-line signature verification system developed within the Laboratoire d'imagerie, de vision et d'intelligence artificielle (LIVIA), as a reference system for our project.

      - Investigate and compare different techniques for features extraction that helps to reduce the watermark size but also ensure the signature verification with low false rejection rate and false acceptances rate.

      - Once the signature verification system established we develop a strategy of encoding and decoding of the watermark and error correction system to minimize the distortion between the decoded and encoded signature.

      - Study the impact of watermarking attacks on the performance of signature verification system.


These steps will allow us to develop a reliable system for gray scale images authentication by handwritten signature.