CHAPTER 2
Literature Review
In this chapter, we discuss the reported literatures about watermarking in image authentication. In section 2.1, we introduce the general watermarking model in image authentication. The categories of watermarking technique are delineated in section 2.2. In section 2.3, we describe the fundamental requirements of watermarking in image authentication. In section 2.4, the categories of malicious attacks are discussed. Finally, we review the various applications of fragile watermarking schemes in section 2.5.
2.1 General Watermarking Model in Image Authentication
In this section, we describe the general watermarking concept in image authentication. Three terminologies used through whole dissertation are introduced at first. “Watermark” is a kind of sensitive messages which are embedded into image to protect the integrity of images. The action of embedding watermarks into images is “watermarking”. After watermarks are embedded into the cover image, the stego-image which contains watermarks is so-called “watermarked image”.
The general watermarking model for image authentication is shown in Fig. 2.1.
The side information for the embedding process is exploited to generate and embed
watermarks. The side information may include the content of original image, watermarks, watermark insertion positions, or secret embedding keys. Watermarks may be encrypted by private-key or public-key systems to form cryptographic watermarks, and then the watermarks should be perceptually embedded into the cover image. After the watermarks are embedded, the watermarked image should be perceptually identical to the original one under normal observation.
The watermarked image may be transmitted to an authorized user through a mistrustful network. After the authorized user received a watermarked image, the authenticity of watermarked image can be evaluated with the side information. The side information exploited in authenticating process may include the original image, the correctly secret keys, or secure information in the embedding process. Supposed
Mistrustful Network
Stego-image
Fig. 2.1 The general watermarking model for image authentication Side Information
Tampered Image
Wrong Key
protected image is tampered, the tampered areas should be accurately localized.
2.2 The Categories of Watermarking in Image Authentication
In recent years, many watermarking researches had been proposed to protect the copyright or integrity of images. According to the various characteristics of watermarking techniques, watermarking schemes can be classified into various categories.
2.2.1 Private, Semi-Private and Public Watermarking
According to the side information which is exploited to extract and verify the embedded watermarks, watermarking techniques can be categorized into three categories: private, semi-private and public [Pet99]. In private watermarking schemes [Cox97][Pod98], when the authenticator received a watermarked image, the original image is indispensable to extract watermarks from the protected image. The private watermarking schemes are also called the non-blind watermarking schemes.
This kind of watermarking can be further categorized into semi-private watermarking technique [Che01][Lu01]. The semi-private watermarking means that the embedded watermarks can be extracted without original image but the original watermarks still need to be acquired from the author. Private watermarking schemes are restricted to be used by the authorized users or content owners.
On the other hand, the public watermarking schemes are also called the blind
watermarking schemes [Che03][Tan04a]. The original image and watermarks are
not required in extracting and authenticating process. No extra cost is required to
send original image and watermarks to authenticator when authorized users want to
verify watermarked image. It is more suitable to be applied in practical applications.
Hence, there is a tendency toward the public watermarking scheme in recent researches.
2.2.2 Spatial and Frequency Domain
According to the embedding strategies, watermarking techniques can be categorized into two categories: spatial and frequency. In watermarking techniques of spatial domain, the watermarks are often embedded watermarks in the nonessential part of the spatial image. Compared to embedding watermarks in frequent domain, the major advantage of embedding them in spatial domain is that the visual quality of watermarked image is easier to be controlled.
On the other hand, after images are transformed into frequency domain, the watermarks are perceptually embedded into frequency coefficients. The major advantage of embedding watermarks into frequency domain is that the distortions made by image processing, such as lossy compression, are easier to be analyzed and controlled. It follows from what has been said that this kind of techniques is more robust to image processing.
2.2.3 Symmetric and Asymmetric Watermarking
According to the encryption method of cryptographic watermarks, watermarking techniques can be categorized into two categories: symmetric and asymmetric watermarking.
In symmetric watermarking or private-key watermarking, symmetric algorithm
uses the identical secure keys for embedding and extracting watermarks. Because
the secure embedding keys must be known by every authorized user to extract
watermarks, the applications of symmetric systems are restricted and this kind of system presents a security risk
In asymmetric watermarking or public-key watermarking, asymmetric algorithms use secure keys for encrypting watermarks during embedding process and public keys for extracting watermarks during authenticating process. In this kind of technique, the secret embedding keys shouldn’t be possibly derived from the public ones. This kind of watermarking method is more suitable for practical applications.
From these remarks one general point becomes very clear: it is a tendency toward asymmetric watermarking in recent researches.
2.2.4 Complete, Soft, and Content-Fragile Authentication
According to the applications of watermarking schemes, the watermarking schemes fall into three categories: complete, soft, and content-fragile [Dit01][Lin00][Zhu03]. The basic idea of complete authentication, also called fragile watermarking technique [Li04b][Tze03], is to embed sensitive messages into an image, so that any modification of pixels of an image will also alter the embedded watermark. Therefore, this kind of techniques can verify the authenticity of protected image and localize the tampered areas through authenticating the distortions of embedded watermark.
A soft authentication, also called semi-fragile watermarking technique [Tan04b][Ye03], is designed to verify the authenticity of protected image even if the protected image is undergoing incident manipulations, such as lossy compression.
In other words, a semi-fragile watermark is capable of distinguishing incident
manipulations from malicious modifications. Thus, this kind of technique provides a
soft authentication for a protected image. The predefined incident manipulations of watermarking techniques also vary according to applications.
The content-fragile watermarking [Dit99][Liu04] is designed to authenticate multimedia content in a semantic level. In other words, the embedded watermarks are authenticated, only if the viewers still have the identical image impression. A content authentication technique extracts the content features of image, and embeds the content features into the regions of interests by applying robust watermarking schemes. The aim of content-based authentication schemes is to reject all manipulations that modify the contents of interests, and to accept all other manipulations even including perceptible manipulations, such as filtering, color manipulation, geometric distortion, etc.
2.3 Fundamental Requirements of Watermarking Scheme in Image Authentication
The desirable features of watermarking scheme in image authentication can be defined as follows [Lin99][Zha04][Zhu03]:
1. Fragility: When images are protected by embedded watermarks, the embedded watermarks should be sensitive to unauthorized manipulations.
2. Identification of Tampered Regions: If a protected image undergoes unauthorized manipulation, then the tampered areas should be accurately identified in the authentication process.
3. Imperceptibility: The difference between the original image and the watermarked image should not be visually perceptible.
4. Security: It should not be possible to guess or counterfeit the embedded
watermarks without knowing the secure embedding key used in the watermark
insertion process, even if the embedding function is known.
5. Noninvertibility of the Watermark: The embedded watermarks should not be invertible without knowledge of the secure embedding key, even if both the watermark generation function and the embedding function are known.
6. Uniqueness: The embedded watermarks should be strongly dependent on the image content, so that even similar images would result in considerably different watermarks.
7. Blind Extraction: The embedded watermarks in a protected image should be blindly extractable only using the secure detecting key.
8. Color Protection: The watermark algorithm should be extendable to color images with any color alteration accurately identified in the authentication process.
9. Computational Efficiency: The watermark embedding and authentication algorithms should be computationally efficient and not overly complex.
10. Authentication Convenience: Authentication of the watermark should require as little information as possible.
2.4 The Classification of Malicious Attacks
The aim of authentication systems is to identify any unauthorized alteration to protected media. There are four general categories of attack in cryptography:
interruption, interception, modification, and fabrication [Sta02]. As in cryptography, several attacks discussed against watermarking-based authentication systems in literatures [Fri02a][Zha04][Zhu03] try to violate the integrity or authenticity of the protected image without being verified by authentication algorithm. The classifications of malicious attacks are shown as:
1. Undetected modifications: This is an attack on integrity. An unauthorized party
may try to make various changes to the protected media without being detected by the authentication system or not to be detected with a “reasonable” probability.
2. Information leakage: An attacker may try to deduce some private information, such as the watermark positions, deriving portions of look-up tables, or the pairs of authentication keys.
3. Cropping attack: Cropping is probably the simplest form of image manipulation.
The image synchronization is easily destroyed after parts of image are cropped.
The unaltered areas should be still verified when the cropping attack happens.
4. Swapping attack: This is a kind of fabrication attack on authenticity. The unauthorized party may swap pairs of watermarked blocks within a watermarked image or from watermarked images. This attack can be regarded as the insertion of spurious message into a protected image. Supposed block-independent watermarking schemes embed watermarks into non-overlapping blocks using the identical embedding keys, the swapped-watermarked blocks are very hard identified. Thus, it can be seen that swapping attack is successfully against block-independent watermarking scheme.
5. Counterfeiting attack: the attack is proposed by Holliman and Memon [Hol00]
and improved by Fridrich et al. [Fri02b]. It is a kind of fabrication attack and successfully against block-independent watermarking scheme. The concept of counterfeiting attack is shown as Fig. 2.2. The aim of counterfeiting attack is that forging watermarks into an unwatermarked image using a watermarked image database. For example, suppose all images in a database were embedded by block-independence watermarking scheme with the identical secure keys, a watermarked image X ~
in database can be expressed as non-overlapping blocks
X ~ i
X n
X X
X ~
||
...
~ ||
~ ||
~
2
= 1
) , (
||
...
||
) , (
||
) , ( ) ,
( 1 1 2 2
2
1
