1. 南台科技大學 資訊工程系 Data hiding based on the similarity between neighboring pixels with reversibility Author:Y.-C. Li, C.-M. Yeh, C.-C. Chang. Date:2012-12-19 Speaker: Xian-Lin Hong Digital Signal Processing, vol. 20, no. 4, pp. 1116–1128, 2010.

2. 2 Outline Problems & Rationale 1 Purpose & Specific Aims 2 3 Materials & Methods 4 Experiments and Result 5 Conclusions

3. 3 1.Problems & Rationale  In recent years, the development of multimedia and computer networks has resulted in the widespread use of digital media to replace traditional postal mail.  Several researchers have employed data compression technology for efficient and robust media transmission via the Internet.  The transmission of digital media in an open Internet channel has increased the risk of incurring leaks of sensitive information.  Protection of sensitive data from attackers in an Internet environment has become an important issue.

4. 4 2.Purpose & Specific Aims  To satisfy the requirements of reversible data hiding and to improve the hiding capacity, this study proposes the difference scheme.  Some applications require high-precision media to recover the original cover image.

5. 5 3. Materials & Methods  The NSAS method utilizes the peak point and zero point-pairs of an image histogram and slightly modifies the pixel values to embed data.  The quality of the stego-image is apparent in that the stego-image has a peak signal-to-noise ratio (PSNR) of at least 48 dB.  The Barbara image with a size 512×512 and 256 gray-level is used as an example to illustrate the NSAS method.  The method includes two processes, i.e., data embedding and data extraction

6. 6 3. Materials & Methods Data embedding process step 1:  For a given image, produce the histogram of the cover image. Fig. 1 shows the histogram of the Barbara image.

7. 7 3. Materials & Methods Data embedding process step 2:  Find and store the most frequent and least frequent pixel values.  For example, in Fig. 1, the pixel value 159 occurs 2576 times, denoted as h(159) = 2576, which is the maximum number of occurrences, and is called the peak point  The pixel value 254 does not appear in the Barbara image, which includes the minimum pixel number, 0, called the minimum point or zero point.  If there is no zero point, make a minimum point by cleaning the data and store the pixel information.

8. 8 3. Materials & Methods Data embedding process step 3:  Scan the cover image once in a sequential order.  If PP > ZP, then shift each pixel value in the range, [ZP+1, PP−1], to the left-hand side by decreasing the pixel value by one unit.  If PP < ZP, then shift each pixel value in the range, [PP + 1, ZP − 1], to the right-hand side by increasing the pixel value by one unit.  For example, in Fig. 1, PP < ZP since 159 < 254, each pixel value in the range, [160, 253], is increased by one.

9. 9 3. Materials & Methods Data embedding process step 3:  As shown in Fig. 2, the histogram changes to make the point, 160, empty. This generates free space to embed data.

10. 10 3. Materials & Methods Data embedding process step 4:  Scan the whole image once again in the same sequential order to embed data.  After scanning the pixel with the peak point value, embed a bit of the hidden data.  If the embedded bit is “1”, then shift the pixel value from PP to ZP by one; otherwise, the pixel value does not change.  Given the Barbara image, if NSAS selects just one pair of peak point and zero point, NSAS can at most embed 2576 bits.  NSAS can select multiple pairs of peak points and zero points to increase its embedding capacity.

11. 11 3. Materials & Methods Data embedding process step 4:  Fig. 3 shows an example of the histogram after data have been embedded.  PP < ZP since 159 < 254.  The pixel value, 159, becomes 160 to embed a bit “1”.  Each pixel in the range [159, 160] can embed a bit.

12. 12 3. Materials & Methods Original Image PP ZP 011234 500112 321014 442016 356406 261621 PP = h(1) = 9 ZP = 7 PP < ZP PP ZP 011345 600113 431015 553017 467507 371731 Shift Image

13. 13 3. Materials & Methods Data extraction process  Step 1: Obtain the peak point and the zero point from the stored record.  Step 2: Scan the stego-image in the same sequential order that was used in the data embedding process.  If the pixel value is in the range [PP+1, ZP], decrease the pixel value by one to recover its original value.  At the same time, a bit "1" is extracted if the pixel value is P+1; a bit "0" is extracted if the pixel value is PP.  Step 3: Restore the corresponding pixel values if the extracted data include the overhead bookkeeping information, which is stored in Step 2 of the data embedding process.

14. 14 3.Background & Literatures Review  PP = h(1), PP < ZP  Secret data :010110110 011345 600113 431015 553017 467507 371731 Shift Image 012345 600123 432015 553027 467507 372731 Cover Image

15. 15 3.Background & Literatures Review Data extraction process  PP = h(1), PP < ZP 012345 600123 432015 553027 467507 372731 Cover Image 011345 600113 431015 553017 467507 371731 Shift Image Secret data :010110110 011234 500112 321014 442016 356406 261621 Original Image

16. 16 3. Materials & Methods  A natural image has local similarity. Therefore, the difference between adjacent pixels is close to zero.  The difference value of the peak point is zero for the transformed Barbara image, i.e., h(0) = 17,563 is greater than 2576.  The transformation increases the free space available for embedding data.

17. 17 4.Experiments and Result  This experiment employed the peak signal-to-noise ratio (PSNR) to evaluate the image quality. The definition of the PSNR value is as follows:  MSE is the mean squared error between the original image and the modified image. According to Eq.  A high PSNR value means high image quality.

18. 18 4.Experiments and Result  All experiments were performed on an AMD K8 3500+ (2200 MHz) PC with 1 GB of main memory.  All algorithms were implemented in Visual C++ 6.0 running on the Windows XP Professional operating system.  Nine experimental images were used, which involved 512 x 512 pixels with 256 gray levels, as shown in next page.

19. 19 4.Experiments and Result

20. 20 5. Conclusions

21. 21 5. Conclusions

22. 22 5. Conclusions  This study proposes a novel data hiding technique, which improves the hiding capacity and the stego-image quality.  The method concerns the pixel difference and shifting pixel values.  The method not include complicated calculations, the method is easy to implement.

23. 南台科技大學 資訊工程系 Thank You !