This essay has been submitted by a student. This is not an example of the work written by our professional essay writers.
The project was started with the learning process and a study conducted by previous researchers. Literature reviews done at the earlier stage to explain overview of theoretical background of project and review previous project of the research area. This chapter gives an overview regarding the theory that is applied into this project.
2.2 Digital Watermarking Basic Principles
Fridrich. J (1998), in his paper  states all watermarking methods share the same building blocks: an embedding system and the watermark extraction or recovery system. In general, any watermarking scheme consists of three parts:
The encoder (Insertion or embedding algorithm)
The decoder (Verification or Extraction algorithm)
Each owner has a unique watermark to incorporate the watermark into the object. The object also known as original image, cover object or host image. Only the authorized users should gain access to the watermarked data.
The stages of watermarking process which comprises of the embedding, the distribution, the extraction and the decision, are described in the following subsections.
2.2.1 Embedding Stage
In this embedding stage, the original image (H) to be watermarked is preprocessed before embedding a watermark (W). Early watermarking schemes worked in the spatial domain, where the watermark is added by modifying pixel values of the host image. Examples of such techniques are Substitution Watermarking and Additive Watermarking . For the case of embedding in the transform domain, this may involve converting the image to the desired domain such as the Discrete Cosine Transform (DCT), the Discrete Fourier transform (DFT) and the Wavelet Transform (WT) domains. Then, to get the watermarked image, inverse transform is performed.
The watermark to be embedded can be a binary image, a bit stream or a pseudo-random number. The key is used to generate a more secure watermark. The watermark key is private and only the authorized person is known. It ensures that only authorized person are able to detect the watermark data. Figure 2.1 illustrates the encoding process.
Figure 2.1 Watermark Embedding
Mathematically, this can be written as
E (H, W, K) = H* (2.1)
Where E is an encoder function and K is the secret key.
The output is the watermarked data. It is perceptually identical to H and is obtained by performing an inverse transform on the altered transform coefficients.
It is passes through the transmission channel. The digital watermarked product will be transmitted through some ways such as internet, or transmission within pen-drive. The channel for the watermarked data could be lossy, noisy and reliable channel. In the process of transmission and distribution of the watermarked image, this will contribute errors to the watermarked image. All these manipulations on the watermarked image have to be seen as an attack on the embedded information. Thus the received watermarked data may be different from the original watermarked data. Details of the attacks will be described in section 2.6.
2.2.2 Extracting Stage
Figure 2.2 illustrates the decoding process.
Figure 2.2 Watermark Extracting
Let's denote a decoder function as D. D takes a watermarked image H* whose ownership is to be determined and recovers a watermark W* from the image using the secret key (K). This watermarking technique is said to be secure since the key used at embedding, is needed for extraction. It is hard to remove or alter the message from the data without knowing the key. Mathematically, this is written as
D (H*, H, K) = W* (2.2)
Where D is a decoder function and K is the secret key.
The extracted watermark (W*) will be used in the decision making stage. The watermarking system examines the extracted data by evaluating the similarity between the original watermark image (W) and the extracted watermark image (W*) during this decoding stage.
The correlation is calculated between the recovered watermark image (W*) and the original watermark image (W) for each of the pixels. Correlation is used as the similarity measure of two images with an appropriate threshold.
Mabtoul. S, Ibn-Elhaj. E, and Aboutajdine, 2006 in their paper  defines correlation as
Where W is original watermark image and W* is the recovered watermark image
Meanwhile, Krishnan Nallaperumal, R.K Selvakumar, S. Rajapandian, K.ArulMozhi and C.Nelson Kennedy Babu, 2006 said in their journal, a higher correlation indicates the existence of the watermark in that band . It is 1, if the original watermark resembles the extracted watermark and 0 otherwise , which can be represented as
Where c is the correlation of two images, is certain threshold.
The threshold for the decision is set as the mean of the correlation value for all the pixels . The watermark is detected if it is larger than the threshold otherwise watermark cannot be found in the image.
2.3 Watermarking Design Issues
The requirement of a watermarking system strongly relies on the particular applications in which it will be deployed. The main requirements which should be fulfilled of digital watermarking system are imperceptibility, security and robustness , , .
The watermarking design should be perceptually invisible so data quality is not degraded and attackers are prevented from finding and deleting it. It is called imperceptible if the watermarked image is perceptually equivalent to the original watermark information .
Measuring imperceptibility in watermarking is important to measure the quality of the image the decoder stage. Peak Signal-to-Noise Ratio (PSNR) is used to indicate the quality of image with watermarking process. The formula is showed below
Where MÃ-N is image size, 255 is gray level range of image, M (i, j) and M' (i, j) are gray level values at pixel (i, j) of original image and watermarked result image respectively. Mean Square Error (MSE) value is the sum between original image and watermarked image in db unit. Na Li, Xiaoshi Zheng, Yanling Zhao, Huimin Wu and Shifeng Li, 2008 in their researches proves that the bigger the value of PSNR, the better the quality of image .
Watermark robustness explains for the capability of the hidden watermark to survive legal of daily usage or any image processing manipulation from intentional and unintentional attackers. Means that, robust watermark has the availability to withstand various image attacks thus providing authentication and ownership identification.
In order to evaluate the robustness of watermarking algorithm, the original watermark, W and watermarked image W* is calculated with the formula below , , :
Where NC is Normalized Cross Correlation
Besides, the watermarking scheme is also tested by various kind of attack to measure the robustness of the algorithm. The watermarking is tested by geometrical attacks, contrast enhancement, adding noise, etc. A full list of attack with related parameters is recorded in section 2.6
Ton Kalker defines watermarking security as "the inability by unauthorized users to have access to the raw watermarking channel". In other words, watermark security refers to the failures of unauthorized users to alter, to remove, to read or to write the watermark content established by robust watermarking . According to Kerckhoffs, in Ke luo and Xiaolin Tian journal, the security must lie in the choice of the key .
2.4 Existing Image Watermarking Techniques
The classification of watermarking algorithm is done in several view points. One of the viewpoints is based on processing domain spatial domain or frequency domain.
2.4.1 Spatial Domain Techniques
Most of the early researches in digital watermark embedded the watermark in the spatial domain which is straightforward, simple and not costly. Least Significant Bit (LSB) is the easiest technique in the spatial domain. Techniques in spatial domain commonly share the following characteristics:
The watermark information is applied in the pixel domain.
No transforms are applied to the cover object in watermark embedding.
Combination with the cover object is in the pixel domain.
The correlation is calculated between the expected patterns with the received signal.
Generally, spatial domain watermarking techniques are not robust against image processing operations because the embedded watermark is not distributed for the whole image and thus contributes in ease to destroy the watermark .
Least Significant Bit (LSB) Technique
The easiest method of watermark embedding is to embed the watermark into the least significant bits of the cover object. In grayscale image, the most significant bit (MSB) is in the left side and the least significant bit (LSB) to the right of 8 bits of a pixel.
Figure 2.3a shows a pixel having the gray value 130. The idea of LSB is to replace the LSB of a pixel with the watermark. As shown in Figure 2.3b, the value changes from 130 to 131 when the LSB is changed. It is undetectable from human eyes. If alter the bit position more closely to MSB, the image will be distorted more, as described in Figure 2.3c .
Figure 2.3 (a) 8 bit pixels with a value of 130. (b) The value changed to 131 after replacing the LSB. (c) The value is changed to 2 after the MSB substitution
2.4.2 Frequency Domain Techniques
In order to have a more powerful technique, frequency domain techniques is introduced such as DCT domain, DWT domain, DFT domain etc.
Discrete Cosine Transform (DCT) Watermarking Techniques
DCT works by separating images into parts of differing frequencies. Only the most important frequencies that remain are used to retrieve the image in the decompression process . The DCT works by separating images into different frequency bands. Thus, it is much easier to embed watermarking information into the middle frequency bands of an image. The middle bands are chosen to avoid the low frequency band that contained the most important parts of image. Moreover, the middle band is selected without exposes themselves to removal through compression and noise attacks in high frequency band . Then, pseudo-random sequences, such as M-sequences, are added to the DCT coefficients at the middle frequencies as signatures.
Figure 2.4 DCT domain watermarking
Discrete Fourier Transform (DFT) Watermarking Techniques
Fourier transform decomposes image function into a set of orthogonal functions, and can transform the spatial intensity image into its frequency domain .
The main drawbacks of using FFT are it has less ability to withstand JPEG compression and cropping attack . Besides, the loss of time information in a signal by Fourier Transform will leads to the difficulty in processing .
Discrete Wavelet Transform (DWT)
Recently, most of the researchers focus on embedding watermark in wavelet domain because of the property of multi resolution analysis that it provides. The existing wavelet based watermarking techniques are explained below:
P. Ramana Reddy, Dr. Munaga and Dr. D. Sreenivasa (2009) present a robust digital watermarking of images by modifying the frequency coefficients of the image based on the Human Visual System (HVS) of image content. The operation of embedding and extraction of the watermark is done in the frequency domain. Therefore, contributes robustness against frequency- based attacks such as compression and filtering. The watermark is embedding into an image by modifying coefficients of mid frequency bands (LH and HL subbands). The experimental result proves the watermarking scheme applied is highly robust against various attacks such as filtering, compression, Gaussian noise etc .
Jianmin Xie and Qin Qin (2010) proposed a watermarking scheme based on the Discrete Wavelet Transform (DWT). In order to improve the watermark is invisibility; algorithm selection coefficient is in high frequency subbands to add a watermark. The experimental results show that the algorithm research paper is feasible, simple and easy to implement. This technique proved to be more robust than the DCT method .
Krishnan Nallaperumal, R.K Selvakumar, S. Rajapandian, K.ArulMozhi and C.Nelson Kennedy Babu (2006) proposed the image is decomposed into wavelet coefficients and a visual recognizable logo and content based watermark information is embedded in the wavelet coefficients. Embedding the watermark in such pixels makes it possible to use maximum amount of watermark due to human eye insensitivity to high entropy areas . Munesh Chandra and Shikha Pandey (2010) give an overview of watermarking techniques and proposed a visible watermarking algorithm for copyright protection of digital images based on DWT. The advantages are due to good time frequency features and well matching with HVS directives .
Na Li, Xiaoshi Zheng, Yanling Zhao, Huimin Wu and Shifeng Li (2008) in their paper proposed a robust algorithm of digital image watermarking based on DWT. This technique adds binary image watermark into gray image. Moreover, the host image is needed for detecting at decoder stage. Watermarking is embedded in the third class HL subbands so that it has unobtrusiveness as well as the robustness to keep quality of the original image. Furthermore, Arnold Transform is applied to binary image to make watermarking more strongly robust against cropping operation . Akhil Pratap Singh and Agya Mishra presents a robust watermarking technique in the gray sale image. In this paper, gray scale image is found to be simpler than other transform technique .
Ke Luo and Xiaolin Tian (2008) proposed a new robust watermarking scheme based on DWT, where a watermark is embedded into a host image twice in two different frequency ranges to withstand different type of image processing attacks. First, the watermark is embedded into lower frequency coefficients. Next, the same watermark is embedded into the mid- frequency coefficients of the host image again to enhance robustness of the watermark. From the results obtained, the watermarks inserted into the middle frequency and high frequency are commonly less robust to low-pass filtering, JPEG compression and scaling, but are extremely robust with respect to salt and pepper noise and rotation. The low frequency watermarks are typically strong robust with nonlinear filtering such as lossy compression, median filter, etc .
Mohamed A. Mohamed, Mohy El-Din A. Abou-Soud, and Mai S. Diab in their paper said, another technique for watermark embedding is to use the correlation properties of additive pseudo-random noise patterns as applied to an image. A pseudo-random noise (PN) pattern of W(x, y) is added to the host image of I(x, y). According to the equation:
(x, y) = I (x, y) + k *W(x, y) (2.7)
Where k denotes a gain factor and the resulting watermarked image
Increase the value of k, will increasing the robustness of the watermark at the expense of the quality of the watermarked image . Moreover, Jianmin Xie and Qin Qin said in their paper, if k has greater value, the stability is better and the invisibility is worse. On the other hand, if k has smaller value, the invisibility is better but the stability is worse .
Mohamed A. Mohamed, Mohy El-Din A. Abou-Soud, and Mai S. Diab in their paper proposed watermarking schemes based on Haar Wavelet Transform. The Haar wavelet transform are chosen because it is simple and fast. It is exactly reversible without any edge effects. They proposed watermarking scheme embeds the watermark in the LH and HL band. The LH and HL values are modified according to pseudo random Number (PN) sequence for the watermark bit 0. Then inverse DWT is used to get the watermarked image. During watermark extraction process, if the correlation of LH and HL values of watermarked image and the PN sequence is greater than the mean correlation then the watermark bit is set to 0 .
2.5 Discrete Wavelet Transform (DWT) Domain Watermarking
Discrete Wavelet Transformation (DWT) transform discrete signal from time domain into time-frequency domain. The transformation result is the coefficients that are enables spectrum analysis of the signal and also spectral behavior of the signal in time .
DWT has been used in digital watermarking more frequently than other frequency domain techniques such as DCT and DFT because of it special characteristics. This is due to multi resolution characteristics, excellent spatial localization, more accurately to the theoretical models of the human visual system (HVS) as compared to DCT and DFT , , .
The human visual system is related to the perceptual quality and measured according to the sensitivity or sharpness of human eye to see details in an image. Based on research in human perception, it is found that the retina of the eye splits an image into several frequency channels. Each signal in these channels is processed independently. This process is similar to DWT multi resolution decomposition. The multi resolution successive approximation enhances the resolution of an image and enhances the resolution of watermark simultaneously. This benefit allows using higher energy watermarks in regions where the HVS is less sensitive. As a result, embedding watermarks in those particular regions provides to increase the robustness of the watermarking techniques.
2.5.1 Wavelet Decomposition
For 2D images, the wavelet transform is done in both horizontal and vertical directions. Firstly, applying 1D wavelet horizontally to each row of the image giving Low Horizontal (LH) and High Horizontal (HH) and then applying on all the column of the image giving (LL, LH) and (HL, HH) can be computed the 2D transform.
The LL sub band represents the coarse-scale DWT coefficients, while LH, HL and HH sub bands represent the fine-scale of DWT of DWT coefficients. The next coarse scale of wavelet coefficient can be determined by further process the LL sub band, until some final scale of N is reachable. When N is reached, 3N+1 sub band is obtained, consisting of the multi resolution subbands LL, LH, HL, and HH sub bands. Example of an image being decomposed into ten sub bands for three levels is shown in Figure 2.5. , , , .
Figure 2.5 DWT Decomposition of Image
2.5.1 Wavelet Image Reconstruction
From DWT coefficients as mentioned before, the original image can be reconstructed. The wavelet image reconstruction is similar to the inverse of the wavelet decomposition. The original image is obtained by concatenating all the coefficients, starting from the last level of decomposition. This process is continued through the same number of levels as in the decomposition process .
2.6 Possible Attacks on Image Watermarking
The attackers intending at the watermarked images can be classified as unintentional or intentional. The attackers have three strategies to defeat watermark robustness as:
To remove enough watermark signal
To jam the hidden communication channel
To desynchronize the watermarked content
The aim of the attackers is to alter, remove or degrade the effectiveness of the watermark. An attack is said to be successful if the attackers disturb any stage of the watermarking cycle .
In this chapter, different types of attacks will be described in order to test the robustness of watermarking schemes. The most popular classification is summarized in Figure 2.5 , , .
Figure 2.5 Attacks on Watermarks
2.7 Applications of Watermarking
There is a wide variety of applications in watermarking. Several applications are listed below.
Digital watermarks are capable to be used in identifying and protecting the copyright ownership of the content . It also can be used in tracing illegally distribution copies .
Identity Card / Passport Security
In the field of data security, watermarks may be used for authentication, certification, and conditional access such as in identity card and passport security . Information in a passport or ID card can also be included in the person's photo that appears on the ID card.
The insertion of the watermark provides an extra level of security in this application. For example if ID card is stolen and he/she replaces the picture, the failure in extracting the watermark will invalidate the ID card , .
Digital content can be watermarked to indicate that the content cannot be illegally replicated and prevent people from making illegal copies of copyright content , .
Digital watermarks can be used to track the usage of digital content. Each copy of digital content can be uniquely watermarked with metadata specifying the authorized users of the content. Tracking application is used to detect illegal copying of content by identifying the users who fake the content illegally. The watermarking technique used for tracking is called as fingerprinting.
Fingerprint is a real advance in identifying real manufactured objects from fake ones based on digital images of the original product stored in a protected server ,.