1. Difference Expansion Based Reversible Data Hiding Using Two Embedding Directions
Yongjian Hu, Member, IEEE, Heung-Kyu Lee, Kaiying Chen, and Jianwei Li
IEEE TRANSACTIONS ON MULTIMEDIA, VOL. 10, NO. 8, DECEMBER 2008
speaker：王軍証 2009/2/13

2. Outline
INTRODUCTION
ALGORITHM
EXPERIMENT AND DISCUSSION
CONCLUSION

3. INTRODUCTION
Multiple-layer embedding results in some unexpected problems.
Image quality(PSNR) drops greatly.
The new difference image has smaller embedding capacity than its predecessor.
The algorithm can not keep its behavior smoothly because each layer embedding has its own embedding capacity limit.
We propose an algorithm that selects expandable differences under the same selection threshold in two difference images and embeds the payload in two orthogonal embedding directions.

4. ALGORITHM A. Scheme for Selecting Expandable Differences
B. Dynamical Search for Expandable Differences in Horizontal and Vertical Difference Images
C. Data Hiding
D. Data Extraction and Lossless Recovery

5. We further define the following notations. P : Payload
We further define the following notations. P : Payload. C : Embedding capacity. A : Auxiliary data package. Q : Header file. M : 2-D binary overflow location map. M : Bitstream of compressed 2-D binary overflow location map. T : Selection threshold. Z : Zero point. B : Embedded bitstream : Function used for calculating the length of a bitstream.

7. A. Scheme for Selecting Expandable Differences
Histogram-Based Difference Selection and Shifting Scheme: Before embedding, Thodi et al. shifted the left outer region and the right one outwards along the difference axis by at least T+1, respectively.

8. Improved Histogram-Based Difference Selection and Shifting Scheme: use the histogram zero points

9. The construction of a 2-D binary overflow location map M
For a difference in the inner region : If does not satisfy (4), its location in M is indicated with “1”.
For a difference in the outer regions : we first calculate the shifted difference value . If (5) is not satisfied, its location in is indicated with “1”

10. B. Dynamical Search for Expandable Differences in Horizontal and Vertical Difference Images

13. C. Data Hiding
Construction of Header File: QH=F(16bits)+TH(8bits) ZHp((TH+1) ×8bits) ZHn((TH+1) ×8bits) MH (24bits)+ PH (24bits)
Embedding for Blind Data Extraction: The bitstream to be embedded is constructed as B=A∪P=Q∪M∪P. We save original LSBs of both shiftable differences and exchangeable differences in a set Ls. We append the bitstream to B.

14. D. Data Extraction and Lossless Recovery
We extract the LSBs of difference image pixels to resume the embedding parameters.
After data extraction is finished, we can begin the image recovery process.

15. EXPERIMENT AND DISCUSSION

19. CONCLUSION
We introduce a dynamical expandable difference search and selection mechanism to balance the use of differences in two embedding directions.
We also exploit zero points and a payload- dependent overflow location map to improve the histogram-based difference selection and shifting scheme.