Block-Level Search Method

In this section, we describe the algorithms for implementing the block-level search method as follows.

Algorithm 4.3. Hiding comment words in a BMP cover image file for later search.

Input: A given BMP cover image and some comment words to be embedded.

Output: A stego-image S.

Steps:

1. Divide the image into several blocks and each block is of size n×n or m×n, where m is not equal to n.

2. Grab a comment word and calculate the ASCII sum modulus 255 of it.

3. Embed the ASCII sum modulus 255, and the comment word itself in the current block, starting from the upper left most pixel of the block, by applying the 4-bit LSB method to the red channel of the pixel’s color values.

4. Repeat the previous two steps until all the comment words are embedded.

5. Take the resulting image as the desired stego-image S.

As an illustration of the data embedding result of applying the above algorithm, Figure 4.4(a) shows a cover image and Figure 4.4(b) shows the resulting stego-image after embedding the comment “This is a Japanese girl beside the window.”

(a)

(b)

Figure 4.4 Result of hiding comment words. (a) A cover image beauty.bmp. (b) The resulting stego-image.

Algorithm 4.4. Searching an image database for desired images.

Input: A key word used for search and a set of stego-images forming an image database.

Output: Desired images with their comments matching the key word.

Steps:

1. Calculate the ASCII sum modulus 255 of the key word.

2. Grab an image from the image database.

3. Compare the key word and the comment word with the following order:

(1) compare the ASCII sums modulus 255 after extracting the value of the ASCII sums modulus 255 from the 4 LSBs of the pixels of the image blocks;

(2) compare the words letter by letter after extracting the current comment word from the 4 LSBs of the pixels of the image blocks.

As long as the values or the letters compared fails to match, skip to check the next comment word. Return the file name of the image if all of the comparisons of a comment word match. Grab the next image if all of the comment words of the current image fail to match the key word.

4. Repeat Steps 1, 2, and 3 until every image has been checked.

5. If no desired image exists, return the message of ‘no such file.’

4.3 Experimental Results

In this section, we show some experimental results. We show a result of non-block-level method first and then that of the block-level one. For the first experimental result of the non-block-level method, we embed different comments into a particular image. The smaller the difference is, the fewer times we need to judge

whether this image is desired or not. The particular image we use is Lena.bmp (shown in Figure 4.5) with its size 512×512 and the first comment we embed is “Lena is a model.” And the second comment we embed is “The girl who lives in the U.S. and studies computer science is called Lena Wang.” Then we use “Lena” as the key to search. Note that the first comment has “Lena” as the first word while the second comment has “Lena” as the second last word. The time elapsed using the first comment to search for the desired image is 0.042153 seconds while the time elapsed using the second comment to find out the desired image is 0.061118 seconds, which is less than a 0.1 second.

Figure 4.5 The particular image, Lena.bmp.

For the second experimental result of the block-level method, we tried to find two desired images (Lena1.bmp and Lena2.bmp) from an image database containing 200 images. Some of the images in the database are shown in Figure 4.6(a). The key word is the same as the previous one (i.e., the key word is Lena) and the comments we embed in the two images are again “Lena is a model.” and “The girl who lives in the U.S. and studies computer science is called Lena Wang.”, respectively The times elapsed to find out the two desired images are around 0.04842 and 0.1046 seconds, respectively. Figure 4.6(b) shows the search results, including the two images with the

key word ‘Lena.’

Lena1.bmp Lena2.bmp AMIGO001.bmp AMIGO002.bmp

AMIGO003.bmp AMIGO004.bmp AMIGO005.bmp AMIGO006.bmp

AMIGO007.bmp AMIGO008.bmp AMIGO009.bmp AMIGO0010.bmp

AMIGO0011.bmp AMIGO0012.bmp AMIGO0013.bmp AMIGO0014.bmp

(a)

Figure 4.6 Image database used in the experiment. (a) Some images in the database and the first two images are Lena1.bmp and Lena2.bmp with comments “Lena is a model” and “The girl who lives in the U.S. and studies computer science is called Lena Wang.”, respectively. (b) Two desired images found in 0.04842 and 0.1046 seconds, respectively.

Lena1.bmp Lena2.bmp (b)

Figure 4.6 Image database used in the experiment. (a) Some images in the database and the first two images are Lena1.bmp and Lena2.bmp with comments “Lena is a model” and “The girl who lives in the U.S. and studies computer science is called Lena Wang.”, respectively. (b) Two desired images found in 0.04842 and 0.1046 seconds, respectively. (continued)

4.4 Discussion and Summary

In this chapter, two methods for fast search of image databases have been proposed. We hide comments inside images so that we do not need additional space to save the comments and as shown in the previous section. Although we can find out desired images in a short time by using either one of the methods, we still feel that there is space to improve the efficiency of both methods. To make the search faster for both methods is our future work.

Chpater 5

Conclusions and Suggestions for Future Works

5.1 Conclusions

In this study, we have proposed several methods for different purposes, including a data hiding method for GIFs, a data hiding method for use on web pages, and two methods for fast image searches. All of them are implemented in a PC environment.

The proposed data hiding method for GIFs can be used to hide huge amounts of data in a GIF file without distorting the image. For the data hiding method used on web pages, we pile up two gray-level PNG images and hide secret data in the alpha channel of the foreground image. And then we adjust both the intensity of the foreground image and that of the background image such that the intensity that the pixel presents after embedding will not be too far away from the original intensity. In this way, we can hide secret messages in a web page without being detected easily.

As to the two methods for fast searches of images in databases, one method is a non-block-level method while the other is a block-level method. Both methods use the comments hidden in images as indices to search and both can find out desired images in a very short time.

5.2 Suggestions for future works

Several suggestions for future researches are listed as follows.

1. Developing a distortion-free and high-capacity data hiding method for GIF images, such that the embedded secret messages are not easily detected by an attacker.

2. Developing high-capacity data hiding methods on other image formats.

3. Developing a data hiding method, based on the proposed concept of piling up two PNGs, which hides secret messages in the alpha channel and the intensity channel of the foreground image, as well as the intensity channel of the background image without producing serious artifacts in the stego-image.

4. Developing faster search methods for the BMP image file and other image formats.

References

[1] E. H. Adelson, “Digital signal encoding and decoding apparatus,” U.S. Patent 4939515, 1990.

[2] D. C. Lou and J. L. Liu, “Steganographic Method for Secure Communications,”

Computers and Security, Vol. 21, Issue 5, pp. 449-460, October 1, 2002.

[3] Y. K. Lee and L. H. Chen, “High Capacity Image Steganographic Model,”

Vision, Image and Signal Processing, IEE Proceedings-, Vol. 147, No. 3, pp.

288-294, June 2000.

[4] N. F. Johnson and S. Jajodia, “Steganalysis: The Investigation of Hidden Information,” Information Technology Conference, 1998. IEEE, pp.113-116, 1-3 Sep 1998.

[5] M. D. Swanson, B. Zhu, and A. H. Tewfik, “Robust Data Hiding for Images,”

IEEE Digital Signal Processing Workshop Proceedings, pp.37-40, 1-4 Sep 1996.

[6] Y. J. Cheng, C. Y. Yin, D. C. Wu, and W. H. Tsai, “Copyright Protection and Authentication of Image Information in Digital Libraries,” Proceedings of 2001 Conference on Trends and Perspectives in Digital libraries and Digital Museums, Hsinchu, Taiwan, Republic of China, pp. VIII 1~11 (in Chinese).

[7] Chen and Lee, “Non-Photorealistic 3D Rendering in Chinese Painting Style,” M.

S. Thesis, Department of Computer Science and Information Engineering, National Cheng Kung University, Tainan, Taiwan, Republic of China, July 2003.

[8] T. K. Yen, “Image Hiding by Random Bit Replacement and Frequency Transformations,” M. S. Thesis, Department of Computer and Information Science, National Chiao Tung University, Taiwan, Republic of China, 1998.

[9] N. Abdulaziz and K. K. Pang, “Wavelet Transform and Channel Coding for Data Hiding in Video,” Info-tech and Info-net, 2001. Proceedings. ICII 2001 - Beijing. 2001 International Conferences, Vol. 2 , 2001, pp. 791 –796.

[10] H. Y. Chang, “Data Hiding and Watermarking in Color Images by Wavelet Transforms,” M. S. Thesis, Department of Computer and Information Science, National Chiao Tung University, Taiwan, Republic of China, 1999.

[11] G. Xuan, J. Zhu, J. Chen, Y.Q. Shi, Z. Ni, and W. Su, “Distortionless Data Hiding Based on Integer Wavelet Transform,” Electronics Letters,Department of Computer Science, Tongji University, Shanghai, China, 2002.

[12] M. Kwan, ^“GIF Colormap Steganography,” 1998.

[13] R. Machado, “EzStego, Stego Online, Stego”, 1997.

[14] H. Repp, “Hide4PGP,” 1996.

[15] A. Brown, “S-Tools for Windows,”1996.

[16] C. Maroney, “Hide and Seek,” 1994-1997.

[17] N. F. Johnson and S. Jajodia, “Steganalysis of Images Created Using Current Steganography Software,”New York: Springer-Verlag, Vol. 1525, Lecture Notes in Computer Science, pp. 273–289, 1998

[18] J. Fridrich and R. Du, “Secure Steganographic Methods for Palette Images,”

Proc. 3rd Int. Workshop Information Hiding, Dresden, Germany, pp. 47–60, 1999.

[19] L. Hongmei, Z. Zhefeng, H. Jiwu, H. Xialing, and Y. Q. Shi, “A High Capacity Distortion-free Data Hiding Algorithm for Palette Image,” Proc. IEEE Int. Symp.

on Circuits and Systems 2, pp. 916–919, 2003.