Advisor：Prof. Chin-Chen Chang Student ：Kuo-Nan Chen Hiding Secret Information in Image Compression Code and Image Protection Techniques 藏匿機密資訊於影像壓縮碼及影像保護技術 Advisor：Prof. Chin-Chen Chang Student ：Kuo-Nan Chen

Motivation Security? Bandwidth? Internet Sender Receiver

A New Data Hiding Strategy with Restricted Region Protection

Introduction (1/3) Traditional information hiding Secret Message: 110110110011 (Cover Image) (Stego Image)

Introduction (2/3) (cover pixels) 00110010 00111100 00111101 01001110 01011010 01011101 01100100 01011111 11001011 10101111 00011110 10010110 10110011 10111100 10011100 01011001 50 60 61 78 90 93 100 95 203 175 30 150 179 188 156 89 (cover pixels) (binary representation of cover pixels) Secret Message: 110110110011 00110011 00111101 00111110 01001111 01011000 01011111 01100100 11001011 10101111 00011110 10010110 10110011 10111100 10011100 01011001 (binary representation of stego pixels)

Introduction (3/3) Our proposed scheme Secret Message: 110110110011 Protected Region (Cover Image) (Stego Image)

Procedure Step 1: Select regions to be protected Step 2: Generate a location map Step 3: Compress the location map with Huffman coding Step 4: Embed information in the cover image

Example 1  changeable pixel 0  unchangeable pixel Cover Image 200 189 170 192 143 155 156 177 45 50 140 163 63 60 75 65 200 189 170 192 143 155 156 177 45 50 140 163 63 60 75 65 1 Cover Image Interesting regions (Step 1) Location map (Step 2)

Add an additional token “01” with frequency = 1 1 (Step 3) Rules for generating tokens Type 1: 11111111 (eight 1s) Type 2: 1…10 (ending with 0) Type 3: 00000000 (eight 0s) Type 4: 00…00 (ending with 0) (Location map) 0001101101111111111111111 Token Frequency 11111111 2 110 000 1 01 Add an additional token “01” with frequency = 1

Condensed location map: 110101000．111 Get the Huffman Code (Step 3) root 1 Token Frequency 11111111 2 110 000 1 01 11111111:2 4 1 110:2 2 1 000:1 01:1 Location map: 0001101101111111111111111 Condensed location map: 110101000．111

Condensed location map (L): 110101000．111 (Step 4) 1 Secret message (S): 11 01 01 01 10 11 01 10 00 11 10 01 11 00 Condensed location map (L): 110101000．111 (Location map) 11001000 10111101 10101010 11000000 10001111 10011011 10011100 10110001 00101101 00110010 10001100 10100011 00111111 00111100 01001011 01000001 11001000 10111101 10101010 11000000 10001111 10011011 10011100 10110001 10011111 00101101 00110001 10001100 10100001 11001011 00111111 00111100 01001011 01000001 (a) the binary representation of cover image, the underlined bits are used to embed L (b) the underlined bits are the result after embedding L to cover image 11001000 10111101 10101010 11000000 10001111 10011011 10011100 10110001 10011111 00101101 00110001 10001100 10100001 11001011 00111111 00111100 01001011 01000001 11001000 10111101 10101010 11000011 10101001 10001111 10011001 10011101 10110001 10011111 00101101 00110001 10001100 10100001 11001011 00111110 00111111 01001001 01000010 00111100 10111111 11000010 10111100 (c) the underlined bits are used to embed S (d) the underlined bits are the result after embedding S to cover image

200 189 170 192 143 155 156 177 45 50 140 163 63 60 75 65 200 189 170 195 169 143 153 157 177 159 45 49 140 161 203 62 63 73 66 60 191 194 188 Cover Image Stego Image

Experimental Results (1/4)

Experimental Results (2/4)

Experimental Results (3/4)

Experimental Results (4/4) where h and w are the height and width of the image, respectively; and and are the cover pixel value and stego pixel value, respectively

A Modification of VQ Index Table for Data Embedding and Lossless Indices Recovery

Introduction Vector Quantization

Vector Quantization (2/2) Decoding

Embedding Procedure

Index Types Type 1: Carry 1 secret bit (no side effect) Type 2: Carry bits with an indicator added in front of it (n = codebook size)

Example Codebook size = 16 Segment of an index table IT Index appearance frequency histogram of IT

high appearance frequency indices Possible Type 1 Indices

Indicators F = R Codebook size = 16  Each index size is 4 bits

Secret bits = 0 1 0 1 1 1 0 1 0 1 Secret bits = 0 1 0 1 1 1 0 1 0 1

Extracting and Recovering Procedure

Experimental Results (1/4) Six 512×512 test images

Experimental Results (2/4) The VQ images compressed by using the codebook sized 256

Experimental Results (3/4) The VQ images compressed by using the codebook sized 1024

Experimental Results (4/4) [1] [1] Z. H. Wang, K. N. Chen, C. C. Chang and M. C. Li, “Hiding information in VQ index tables with reversibility,” Proceedings of the Second International Workshop on Computer Science and Engineering, Qingdao, China, pp. 1-6, October 2009.

Thanks for your listening

Logistic Map

where , and r is a positive number. Chaotic maps Logistic map , where , and r is a positive number. One of the simplest chaotic maps Generate unpredictable results Guite sensitive to their initial conditions (butterfly effect)

logistical map The horizontal axis shows the values of the parameter r and the vertical axis shows the possible long-term values of X.

Hamming Code

Hamming Code R. W. Hamming, “Error detecting and error correcting codes,” Bell system technical journal, vol. 26, no. 2, pp. 147-160, April 1950. The most widely used Hamming code is (7, 4). D1 D2 D3 D4 P1  P2 P3 註：data (D1, D2, D3, D4), parity check bits (P1, P2, P3)

P1 P2 D1 P3 D2 D3 D4 The normal form of the (7, 4) Hamming code D1 D2 D3 D4 P1 P2 P3 The reorganized form of the (7, 4) Hamming code used in our proposed scheme (a) (b)