A Hybrid Digital Watermarking Approach Using Wavelets and LSB

ABSTRACT


INTRODUCTION
In recent years, increase in use of the widespread internet has allowed the authors to distribute their content in digital form. Digital watermarking is the one of main richest research topic in the field of image processing and it has great attention for research community. Compare to audio and video, image watermarking is more dedicated and popular because of many practical applications such as Authentication of content and objects, Content identification and management, copy right protection and so on. The content scheme. However, it is impossible to put all the applications in one scheme because different applications demand different properties of watermarking system to different extent. Depending on the watermarking applications and purpose, different properties or requirements of watermarking also arise and result in various design issues.
To overcome the disadvantages, the present paper proposes a method called Wavelet based Least Significant Bit Watermarking (WLSBWM) integrates the alphabet pattern approach for generating the shuffled image, wavelet concept to reduce the dimensionality, Pell's cap map for protection from attacks and LSB approach is used to insert the watermark image. The present approach is simple technique to insert the image and provides high protection from attacks. The novelty of the proposed approach is that double protection is provided for watermarked image so that it protect from attacks. The rest of the paper is organized as follows. Proposed WLSBWM described in section 4 and results are discussed in section 5. Attacks on the proposed method are discussed in section 6 and finally conclusions are given in section 7.

PROPOSED METHOD
In order to provide copyright protection for the identification of ownership, the present paper provides a hybrid technique to insert and extract the watermark in effective and efficient manner. The proposed WLSBWM method consists of 8 simple steps for inserting the watermark image and 8 steps for extracting the watermark image. The block diagram of the inserting water mark image is shown in Figure 1. The watermark insertion algorithm is described below.

A. Watermark insertion algorithm
Step 1: Identify the Alphabet pattern: In insertion algorithm step one, for providing the security to protect from attacks the present approach converts the original image into shuffled image. The present paper uses the Alphabet patterns to generate the shuffled image. The generation of shuffled image has two sub tasks i.e. identify the Alphabet patterns on each 3×3 and change the direction of the pixel values in reverse direction. The present paper uses 'T' pattern, 'E' pattern, and 'U' patterns. The 3×3 window consists of 9 pixels. The pixel values are indicated by P 1 , P 2 , P 3 … P 9 . The 3×3 window is shown in Figure 2. P 1 P 2 P 3 P 4 P 5 P 6 P 7 P 8 P 9 Figure 2. 3×3 window In a given window, if the pixels values of P 1 , P 2 , P 3 , P 5 , and P 8 are same then treats the3×3 window forms the 'T' pattern. If 'T' pattern existed in 3×3 window then change the direction of the pixel positions to form inverted T pattern. The figure 3 depicts the inverted 'T' pattern. If the pixel positions shown in figure 4 which are highlighted has same values then 3×3 window forms the E pattern and change the pixel position according to figure 4(b). In the same way if the 3×3 window form U pattern, change the positions of the pixels according to figure 5b. The same procedure is applied for remaining 3×3 windows in the entire image the resultant image is treated as shuffled image  Figure 3. (a) T Pattern (b) Inverted T pattern P 1 P 2 P 3 P 3 P 8 P 1 P 4 P 5 P 6 P 6 P 5 P 4 P 7 P 8 P 9 P 9 P 2 P 7 (a) (b) Figure 4: (a) E Pattern (b) Inverted E pattern P 1 P 2 P 3 P 7 P 8 P 9 P 4 P 5 P 6 P 4 P 5 P 6 P 7 P 8 P 9 P 1 P 2 P 3 (a) (b) Figure 5. (a) U Pattern (b) Inverted U pattern Step 2: Pell's Cat Map (PCM): For providing further security and authentication, Pell's Cat Map (PCM) [37] is employed on the 5×5 non overlapped blocks of shuffled image.
A discrete mapping using the matrix P = with determinant −1 is still area preserving but also orientation reversing. As it turns out the matrix P will generate numbers in the Pell's and half-companion Pell sequences, so Ptogether with the modulo N operation will henceforth be denoted Pell'scatmap as shown in equation (1) (1)

Where
Step 3: Apply DCT on mixed image before inserting the watermark image. DCT has been extensively used in image watermarking because of high energy compaction competence and respectable robustness. Generally, from spatial domain to frequency domain conversion Discrete Cosine Transform (DCT) is used [38,39]. It also delivers suitable trade-off between Human Visual System (HVS) model and the image misrepresentation degree [40,41]. DCT watermarking can be classified into two categories: Global DCT watermarking and Block-based DCT watermarking [42,43]. The DCT computation is performed on the entire image in Global DCT [41], whereas the DCT computation is performed separately on each non-overlapping blocks [44,45] to get low-frequency, mid-frequency and high-frequency subbands [43]. Generally, the watermark is inserted into a mid-frequency sub-band, which provides protection from watermarking attacks and it is well-matched with HVS model [46,47]. Given an image f of size M x N, the forward and inverse DCTs are shown in equations (2) and (3) [48]. The present paper utilizes and applies DCT on mixed image.

Step 4: Apply N level DWT on DCT image:
In frequency domain, another reliable transformation technique is Discrete Wavelet Transform (DWT). DWT is a mathematical tool for disintegrating an image hierarchic [39]. It divides the image into four sub-bands which are lower resolution approximation image (LL), horizontal (HL), vertical (LH) and diagonal (HH) detail sub-bands [37]. This process of division can be repeated several times to compute multi-level wavelet decomposition. Based on HVS model, the LL sub-band is not suitable for the watermark embedding, because it contains important data about the image and causes image distortion. HH sub-band is not suitable because of less hearty against image processing operations such as lossy compression [45]. Thus, the suitable subbands for watermark embedding are the mid-frequency sub-bands LH and HL [46,48]. Figure.6 illustrates decomposition of an image using 2D wavelet transform after 3 levels of decomposition. Apply N th level DWT on DCT image to insert the watermark, N level depends on the Size of the original image and water mark image. Suppose the size of the image 256×256 and the watermark image size is 64X64 the 2 level DWT is applied. If the size of the original image is 512×512 then 3 levels of the DWT applied on original image.
Step 5: Embedding the watermark: Find the Size of the Watermark image and Converts the watermark image into a vector of zeros and ones. The condition for inserting the watermark is the size of the LHn is equal to size of the watermark image. Where n is the n th level DWT. The LSB of the each value in LHn sub-band is replaced with the corresponding watermark image bit value. The new LHn sub band is called the watermark sub band image Step 6 Apply N th inverse DWT: Apply nth level Inverse DWT on watermark sub-band image and IDCT is also applied.
Step 7: The reverse of step 3, inverse PCM is applied on the shuffled watermarked image.
Step 8: The reverse of step one, Identify the Alphabet patterns on each 3×3 of shuffled watermarked image and change the direction of the pixel values in reverse direction to obtain a shuffled image. The resultant image is called watermarked image.

B. Water mark extraction algorithm
The block diagram of the water mark extraction is shown in Figure 7. The proposed method Wavelet based LSB Watermark Extraction (WLSBWME) consists of 8 steps as illustrated below.
Step 1: In step one, Identify the Alphabet patterns on each 3×3 sub-window of the watermarked and change the direction of the pixel values in reverse direction to obtain a shuffled watermarked image.
Step 2: Apply Pell's Cat Map (PCM) on the each 5×5 sub-window of shuffled image.
Step 3: Convert the watermark image into a vector of zeros and ones and find the Size of the Watermark image Step 4 & 5: Apply DCT on watermarked shuffled image and get watermarked shuffled DCT Image Step 6: Apply N level DWT on DCT image to extract the watermark image.
Step 7: After N th DWT is applied on image, stores the LH1 values into S.
Step 8: extract the LSB of the each values in S, store the values inTemp which is equal size of the S. The Temp is the watermark image.

RESULTS AND DISCUSSION
The proposed WLSBWM method is experimented over 30 images of size 256×256. The images used in this approach are shown in figure 8. The present method is tested with two different watermark images. i.e. 'AITAM' and 'GITAM' logos of size 64×64and shown in figure 9(a) and 9(b) respectively. The proposed method tested with Matlab software on i3 processor and 4GB RAM. The resultant watermarked images after inserting the watermark image logos of 'GITAM' and 'AITAM' are shown in Figure 10 and 11 respectively   To find the effectiveness of the proposed method, the present paper used two popular and effective criteria called Normalized Correlation Coefficient (NCC) and Peak Signal Noise Ratio (PSNR) for evaluating the performance of the proposed watermarking algorithm.
The quality of the watermark or the frangibility of the algorithm is assessed by the similarity measurement NCC between the referenced watermark W and the extracted watermark W* as given in (4) Where, N is number of pixels, w(i) and w*(i) are the original watermark and the extracted watermark. In the above equation ρ = 1 indicates perfect correlation, while an extremely low value reveals that the watermarks are dissimilar. If NCC value ranges from 0.65 to 1.0 then one can say that the image preserves high quality after inserting the watermark.
The present method also calculates, the difference between the original image and the watermarked image by Peak Signal Noise Ratio (PSNR). The bigger PSNR is, the smaller is the difference, and PSNR is defined through given Equation (5)

MN MSE
Where M and N are respectively the length and the width of the host image; X ij denotes the gray level of the original image pixel; X' ij denotes the gray level of the watermarked image pixel.

PROPOSED WLSBWM METHOD WITH ATTACKS
To find the effectiveness of the proposed WLSBWM method, find the two parameters values when attacks on the image. Watermarking techniques are usually tested against various robustness criteria.
The proposed watermarking technique is tested by using the different geometric attacks and transformations on Barbara, Monalisa, MR-1 and Eagle-1. The resultant watermarked 'Barbara' with different attacks like salt and pepper noise, rotation, median filter, cropping, Gaussian noise, compression, Grey level blurring, Motion blurring and Sharpening are shown in Figure 12 (a) to 12( i).   Table 3 and 4.

Comparison of the proposed method with other existing methods:
To evaluate the efficiency, the proposed method is compared with the existing watermarking approaches [46,47,48]. The method proposed by Zhu Yuefenget.al [46] inserts and extracts the water mark image using dual transformation and self-recovery approach. This approach analyzes inserting positions by using DC coefficient and inserts the water mark image into original image. Saravjit Kaur [47] proposed water marking technique based on DWT. The insertion and extraction of the watermark image in the grey scale images are accomplished by transform methods. Thirugnanam et.al. [48] Proposed a technique using DWT and Independent Component Analysis (ICA). The performance results of the proposed WLSBWM method and other existing methods are listed in table 5. Table 5 clearly indicates the WLSBWM method outperforms the other existing methods. The graphical representation of the performance of the WLSBWM method and other existing method is shown in Figure 13.  Figure 13. Graphical Representation of the proposed WLSBWM with the existing methods.

CONCLUSIONS
The present paper derived a hybrid scheme called WLSB for embedding the watermark. The proposed scheme uses three stages for embedding the watermark. In the first stage, shuffled image is generated by using alphabet patterns and PCM for protection from attacks. In the second stage, the DCT is applied and then N level DWT is applied until size of the LH1 sub band size matches with water mark image. Insert the watermark bits into LSB of the LH1 sub band values row by row and column by column. The proposed scheme guarantees high authentication. The present approach is simple and reliable and provides more security. The extraction process is also handy with simple steps. The experimental results on various images with various attacks show that the proposed technique provide good image quality and robustness when compared to other methods.