SELF-AUTHENTICATION OF COLOUR IMAGES I. Kostopoulos1 , A. N. Skodras2 , and D. Christodoulakis3
1
Computer Technology Institute R. Fereou 61 GR-26221 Patras Greece
[email protected]
2
Electronics Laboratory University of Patras GR-26110 Patras Greece
[email protected]
3
Computer Engineering & Informatics Department University of Patras GR-26110 Patras Greece
[email protected]
ABSTRACT Digital watermarking along with recent cryptographic techniques are today’s most promising methods for copyright protection, authentication, labeling and monitoring of digital media. In this work a semi-fragile watermarking scheme is presented. This scheme is applied to digital images that represent pieces of art, like scanned paintings published in the web. The method aims at the authentication of a specific region of the image, which in the case of scanned paintings is the author’s signature. In this manner we extend the idea of embedding a binary logo for authentication purposes, by embedding a piece of information of the image itself (self authentication).
1. INTRODUCTION The term ‘authentication’ has a wide range of meanings, as for example that of a specialist that decides whether a piece of art is authentic or not, whether a user can view or download that piece of art and finally the decision as to whether the content of an object is staying intact after its publication or transmission on the Internet. Many cultural organizations spend time and money investing on new technologies of image archiving, digital library construction etc. At the same time these organizations can guarantee the authenticity of the pieces of art they possess, since they have both the ownership and the experts opinions. From the moment that these works of art are digitized and published on the Internet, several problems start arising. It is not unusual to find on the Internet several digital images that have many differences, but at the same time pretending to represent the same piece of art. The proposed solutions for image authentication follow two basic directions. The first direction is the encryption, where the data representing the image along with a secret key are given as an input to a secure one-way hash function producing an output message that is used for checking the data integrity. These cryptographic methods are asymmetric, since a different key called public key is employed next together with the hashing product to determine if the data have been altered or not. The problems of such kinds of methods are twofold: firstly they use other than the actual data information for the integrity check and
secondly that they are very fragile. Even a bit change, is giving completely different integrity results. The other direction is watermarking. The advantages in the case of watermarking methods are that the authentication information is embedded into the host data and at the same time they are not as fragile as cryptographic ones. An early attempt in the field of image authentication was the method proposed in [1] where the checksums of digital images were calculated and in combination with a seal produced the watermark information that was responsible for the authentication. This work excited the idea of digital image authentication and many researchers approached the problem from different and more sophisticated ways. Another method that is both efficient and easy to compute, was proposed by Yeung and Mintzer [2]. According to that method a binary logo is embedded in an image in order to detect possible alterations in the image and at the same time provide some information about the image owner. A general overview of the digital watermarking systems and methods is given in [3], while a more specific overview of a method designed with the purpose to detect the alterations in digital images (tamper proofing) is given in [4]. More recent methods [6,7] propose the embedding of an approximation of the image in the two least significant bits of the original image or a decision criterion about the maliciousness compression of the image. If the watermarked image is found altered then they can recover the lost information. However, the first method is very fragile since modifications of the two least significant bits cause severe damage to the watermark information, while the second method aims to recover every part of the image and not specific ones. The present work belongs to the class of methods that embed a logo in the image once or multiple times in order both to authenticate the owner of the image and to detect the possible alterations of it. Instead of the binary logo of the company or organization a piece of the image itself is embedded in it. The paper is structured as follows: in section 2 the proposed method is presented. In section 3 a set of experiments and the corresponding results are given. Finally, in section 4 conclusions are drawn and future directions are discussed.
2. THE PROPOSED METHOD Let I be an MxN colour image and S the region of the image I that represents the author’s signature. Each pixel value of the image is given by the composition of the three colour components R, G, B that are considered as MxN 8-bit matrices. The watermark information is embedded in several pixels of one or more colour components, in a way that 1 bit of information affects not more than 4 of the significant bits (LSB) of each pixel intensity. Let B(i,j), i=0,1,…,M-1, j=0,1,…,N-1 be an 8-bit pixel value of the blue component. Embedding one bit of information into this value imposes the mapping of this value to x in a way that |B(i,j)-x|
