Academia.eduAcademia.edu

Figure 3 – uploaded by Arulmozhivarman Pachiyappan

See full PDFdownloadDownload figure
Table 2 Threshold values of sample cover images Hadamard transformed cover and watermark images are shown in Figures 4(a) and 4(b) respectively. From Figure 4, it is evident, that the energy is concentrated over the highly textured blocks compare to the smooth regions. The same observation has been made for all the images in the cover image database.

Table 2 Threshold values of sample cover images Hadamard transformed cover and watermark images are shown in Figures 4(a) and 4(b) respectively. From Figure 4, it is evident, that the energy is concentrated over the highly textured blocks compare to the smooth regions. The same observation has been made for all the images in the cover image database.

Related Figures (16)

Figure 1 Representation of classified blocks (see online version for colours) The calculated entropy is normalised by dividing it with the maximum entropy. Using Otsu (1979) method, a threshold value is calculated over the normalised entropy values. This is the regular method for threshold selection. It assumes that there are two classes and calculates the threshold separating the two classes so that their intra-class variance is minimal. Using the threshold obtained, each block of an image is classified into low intensity smooth blocks, high intensity smooth blocks, Higher the value of entropy more the information content and the corresponding block could be classified as textured blocks and low entropy value blocks are classified as smooth blocks. Further the blocks are classified into low intensity smooth blocks and high intensity smooth blocks. Similarly the texture blocks are further classified into low texture blocks and highly textured blocks.
Figure 1 Representation of classified blocks (see online version for colours) The calculated entropy is normalised by dividing it with the maximum entropy. Using Otsu (1979) method, a threshold value is calculated over the normalised entropy values. This is the regular method for threshold selection. It assumes that there are two classes and calculates the threshold separating the two classes so that their intra-class variance is minimal. Using the threshold obtained, each block of an image is classified into low intensity smooth blocks, high intensity smooth blocks, Higher the value of entropy more the information content and the corresponding block could be classified as textured blocks and low entropy value blocks are classified as smooth blocks. Further the blocks are classified into low intensity smooth blocks and high intensity smooth blocks. Similarly the texture blocks are further classified into low texture blocks and highly textured blocks.
Table 1 Block classification and the corresponding components used for calculation of scaling and embedding factors For low intensity smooth regions the scaling and embedd ing factors are derived from Hadamard DC coefficients. For high intensity smooth regions the scaling and embedding factors are derived from Hadamard DC coefficients and the Similarly for low textured blocks, the scaling and embedd contrast index of each block. ing factors are derived from Hadamard DC and AC coefficients. Finally for highly textured blocks the scaling and embedding factors are derived from both Hadamard DC and AC coefficients and from contrast index. For each block b;, the contrast index is ca culated using (4)
Table 1 Block classification and the corresponding components used for calculation of scaling and embedding factors For low intensity smooth regions the scaling and embedd ing factors are derived from Hadamard DC coefficients. For high intensity smooth regions the scaling and embedding factors are derived from Hadamard DC coefficients and the Similarly for low textured blocks, the scaling and embedd contrast index of each block. ing factors are derived from Hadamard DC and AC coefficients. Finally for highly textured blocks the scaling and embedding factors are derived from both Hadamard DC and AC coefficients and from contrast index. For each block b;, the contrast index is ca culated using (4)
The scaling factor and embedding factor for low textured blocks and high textured blocks are calculated using (15), (16) and (17), (18) respectively. The architectural diagram of proposed watermarking model is shown in Figure 2. Two separate databases for cover images and watermark images are stored. For a given cover image and for a selected watermark image, the embedding process is discussed below. The cover image is divided into many smaller blocks and they are classified based on the procedure discussed in Section 3.1. All the classified blocks are subjected through Hadamard transformation as given in (3) to derive scaling and embedding parameter as per the procedure presented in Section 3.2. The watermark image to be imprinted is also divided into smaller blocks of size bs x bs and Hadamard transformed using (19).
The scaling factor and embedding factor for low textured blocks and high textured blocks are calculated using (15), (16) and (17), (18) respectively. The architectural diagram of proposed watermarking model is shown in Figure 2. Two separate databases for cover images and watermark images are stored. For a given cover image and for a selected watermark image, the embedding process is discussed below. The cover image is divided into many smaller blocks and they are classified based on the procedure discussed in Section 3.1. All the classified blocks are subjected through Hadamard transformation as given in (3) to derive scaling and embedding parameter as per the procedure presented in Section 3.2. The watermark image to be imprinted is also divided into smaller blocks of size bs x bs and Hadamard transformed using (19).
Figure 2 Proposed watermark embedding model After inserting watermark, inverse Hadamard transformation is applied on each block to reconstruct watermarked image Jj, by rearranging all the blocks. The algorithm to embed watermark in a cover image is given in Algorithm 1.
Figure 2 Proposed watermark embedding model After inserting watermark, inverse Hadamard transformation is applied on each block to reconstruct watermarked image Jj, by rearranging all the blocks. The algorithm to embed watermark in a cover image is given in Algorithm 1.
The obtained threshold value using Otsu method is given in Table 2 which is used for classifying blocks into category of low intensity smooth blocks, high intensity smooth blocks, low textured blocks and highly textured blocks. Figure 3. Sample watermark image
The obtained threshold value using Otsu method is given in Table 2 which is used for classifying blocks into category of low intensity smooth blocks, high intensity smooth blocks, low textured blocks and highly textured blocks. Figure 3. Sample watermark image
Figure 4 (a) F-16 image in Hadamard domain (b) Watermark image in Hadamard domair
Figure 4 (a) F-16 image in Hadamard domain (b) Watermark image in Hadamard domair
Figure 7 shows some sample cover images and the corresponding watermarked images. For the F-16 image which is basically a high contrast image, the visibility of imprinted watermark quality is good. The upper portion of the watermark is embossed in the high intensity smooth region where in the scaling factor and embedding factor calculated adaptively makes the watermark perceptible. The visibility of watermark exists in textured region is poor to leave the underlying content more visible. Figure 6 Results obtained at every step for F-16 image
Figure 7 shows some sample cover images and the corresponding watermarked images. For the F-16 image which is basically a high contrast image, the visibility of imprinted watermark quality is good. The upper portion of the watermark is embossed in the high intensity smooth region where in the scaling factor and embedding factor calculated adaptively makes the watermark perceptible. The visibility of watermark exists in textured region is poor to leave the underlying content more visible. Figure 6 Results obtained at every step for F-16 image
Table 3. Normalised correlation coefficient values for sample cover images Figure 9 shows the comparison of entropy values of the sample greyscale cover images and watermarked images. In the House image alone, the difference is more than 0.1 whereas in all other cases, the entropy values are similar. The reason for the large difference in House image is due to the fact that the cover image is of uniform background and foreground portion is also uniform and hence the textural content is slightly varied.
Table 3. Normalised correlation coefficient values for sample cover images Figure 9 shows the comparison of entropy values of the sample greyscale cover images and watermarked images. In the House image alone, the difference is more than 0.1 whereas in all other cases, the entropy values are similar. The reason for the large difference in House image is due to the fact that the cover image is of uniform background and foreground portion is also uniform and hence the textural content is slightly varied.
Figure 9 Entropy comparison between greyscale cover and watermarked images (see online version for colours) The perceptibility of the watermark image made the textural content slightly distorted when smooth regions in the cover image are predominant. Figure 10 shows the comparison of contrast values between the cover image and watermarked images. The contrast variation is very minimal for all the sample images presented.
Figure 9 Entropy comparison between greyscale cover and watermarked images (see online version for colours) The perceptibility of the watermark image made the textural content slightly distorted when smooth regions in the cover image are predominant. Figure 10 shows the comparison of contrast values between the cover image and watermarked images. The contrast variation is very minimal for all the sample images presented.
Figure 10 Comparison of contrast values between greyscale cover and watermarked images (see online version for colours) The fourth metric used is the energy component. Figure 11 shows the comparison of the energy in cover data and watermarked data. Except for the House image, the energy of the cover data and watermarked data are similar. Finally, the PSNR obtained for the cover image and watermarked image is shown in Table 4. The higher the PSNR values the less is the noisy content. The values obtained shows that the insertion of the watermark content has not introduced much noise in the image.
Figure 10 Comparison of contrast values between greyscale cover and watermarked images (see online version for colours) The fourth metric used is the energy component. Figure 11 shows the comparison of the energy in cover data and watermarked data. Except for the House image, the energy of the cover data and watermarked data are similar. Finally, the PSNR obtained for the cover image and watermarked image is shown in Table 4. The higher the PSNR values the less is the noisy content. The values obtained shows that the insertion of the watermark content has not introduced much noise in the image.
Figure 11 Comparison of energy values between greyscale cover and watermarked images (see online version for colours)
Figure 11 Comparison of energy values between greyscale cover and watermarked images (see online version for colours)
Table 4 Peak signal to noise ratio between cover and watermarked images In Figure 12, the extracted watermark from various watermarked images are shown. The extracted watermark quality is considerable but unauthorised users can not remove the watermark without much effort. At the same time while attempting to remove the watermark the quality of underlying cover data will also be lost and become no longer useful to unauthorised users.
Table 4 Peak signal to noise ratio between cover and watermarked images In Figure 12, the extracted watermark from various watermarked images are shown. The extracted watermark quality is considerable but unauthorised users can not remove the watermark without much effort. At the same time while attempting to remove the watermark the quality of underlying cover data will also be lost and become no longer useful to unauthorised users.
Related topics:
Information SystemsComputer ScienceComputer Networks and Information Security
580 California St., Suite 400
San Francisco, CA, 94104
© 2025 Academia. All rights reserved