Reversible hiding in DCT-based compressed images

1. Reversible hiding in DCT-based compressed images Authors:Chin-Chen Chang, Chia-Chen Lin, Chun-Sen Tseng and Wei-Liang Tai Adviser: Jui-Che Teng Speaker: Gung-Shian Lin Date:2009/12/17

2. Outline 1. Introduction 2. Related works 3. Proposed scheme 4. Experimental results 5. Conclusions

3. Introduction • Lossless and reversible steganography scheme for hiding secret data in each block of quantized DCT coefficients in JPEG images.

4. Introduction • In 2001, Fridrich et al. proposed Invertible authentication watermark for JPEG images. • In 2004, Iwata et al. proposed Digital steganography utilizing features of JPEG images.

5. Quantization Table RGB Image Quantization Transformation RGB→YCbCr Runlength coding Huffman coding Huffman Table Composition MCU JPEG Image 2-D DCT Related works • RGB transformation for JPEG

6. Proposed scheme Embedding procedure bi be the length of ceaseless zeros zi,1 represents the zero value of the lowest frequency R9 R7 R5 R3 R1 R8 R6 R4 R2

7. Proposed scheme si be the secret bit we want to embed into set Ri R5 R3 R4 R2 R1

8. Proposed scheme • The embedding strategies and elimination measures for ambiguous conditions are as follows: Case 1:If bi ≧2, we use the value of zi,2 to indicate the hidden secret bit in set Ri (1≦i≦9). We modify the value of zi,2 to hide secret bit by using the Eq. where 1 or -1 is randomly selected.

9. Proposed scheme • Ambiguous condition A and its remedial measure。 R1

10. Proposed scheme where 3≦(j－1)≦ki R2

11. Proposed scheme Case 2: If bi < 2 and both zi,1 and zi,2 do not exist, none secret bits can be hidden in a set Ri. • Two ambiguous conditions may exist, and therefore two remedial measures for eliminating them are described below. • Ambiguous condition B and its remedial measure • Ambiguous condition C and its remedial measure

12. Proposed scheme • Example of embedding:assume four secret bits, 0, 0, 1 and 1

13. Ri ri,j＝1 or -1 ri,j≠1 or -1 ri,j does not exist ri,j+1＝0 ri,j+1≠0 ri,j-1＝0 and ri,j-2＝0 j≦2 si=0 mark ri,1 as zi,2 ri,j-1＝0 and ri,j-2＝0 j≦2 si=0 mark ri,j-2 as zi,2 si does not exist si=1 mark ri,j as zi,2 si=0 mark ri,j-2 as zi,2 si does not exist Proposed scheme • Extracting procedure Step 5. Repeat Steps 3 and 4 until all blocks are processed. Step 4. Extract si from set Ri by using the following rules: Step 3. For each set Ri in a block, let ri,j be the highest frequency non-zero component, where 1≦i≦9 and 1≦j≦ki. Step 2. Scan each block according to a predetermined order. Step 1. Obtain non-overlapping 8 * 8 blocks of quantized DCT coefficients of the Y components from a JPEG stego-image after Huffman decoding and runlength decoding.

14. Proposed scheme • Restoring procedure • Rule 1: If si exists and r’i,j+3＝0, where 4≦(j＋3)≦ki, then the original value of r’i,j+2 is restored by using Eq. • Rule 2:If si does not exist and the two highest coefficients (r’i,1，r’i,2) of set Ri equals (x, 0), where x≠0, then the original value of r’i,1 is restored by using Eq. where 3≦(j＋2)＜ki.

15. Proposed scheme • Rule 3: If si does not exist and the pair having the three highest coefficients (r’i,1, r’i,2, r’i,3) of set Ri equals (0,x,0), where x≠0, then the original value of r’i,2 is restored by using Eq.

16. Proposed scheme

17. Proposed scheme • Modifying quantization table for better image quality and hiding capacity

18. Experimental results

19. Experimental results

20. Experimental results

21. Conclusions • The scheme provides stego-images with acceptable image quality and similar hiding capacity can be achieved with the Iwata et al. scheme。 • The scheme can withstand visual and statistical attacks 。