The Study of Stegnography and Watermark on Binary

隱像術與數位浮水印在二元影像上之研究 The Study of Stegnography and Watermark on Binary Images 指導教授 ： Jan, Jin-Ke 協同指導教授： Chang, Chin-Chen 研究生 1 ： Wang, (詹進科) (張真誠) Chung-Chuan (王中全) Department of Computer Science and Engineering, National Chung Hsing University

國立中興大學 資訊科學與 程學系 Outline 1 2 3 Related Work Evaluation Proposed Schemes Conclusions 4 5 2 Introduction Future Works

國立中興大學 資訊科學與 程學系 Introduction (1/2)-Information Hiding Sender illegal Binary Image Internet Secret watermark 3 Secret watermark Receiver

國立中興大學 資訊科學與 程學系 Introduction (2/2)-Information Hiding in Binary Images (0 -255) Pixel: (0 -255) (0, 1) Binary image Gray-scale image Imperceptibility Binary (edge) Gray-scale (All pixels ) Color (All pixels ) 4 (0 -255) Color image Capacity Distortion Methods Small ? Normal LSB, … Large LSB, …

國立中興大學 資訊科學與 程學系 Part I - Related Work Evaluation n 5 Imperceptibility (discuss on edge and artifact) n Subjectively : naked human eyes n Objectively : per pixel distortion Capacity (not only size but hiding bits in a pattern block) n Size of embeddable block n Hiding bits per embeddable Reversibility (Both functional and practical)

國立中興大學 資訊科學與 程學系 Imperceptibility Evaluation(1) Not boundary Boundary Non-reversible 1. 2002 Tseng et al. [F K W-4× 5, 3] 2. 2003 Chen et al. [Characteristic value-4× 4, 1] 3. 2004 Wu et al. Artifact [Shuffle-3× 3 least block, 1] 4. 2007 Yang et al. [4 Indices- 3× 3, 1] Reversible 5. 2005 Tsai et al. [PWLC – 1× 6, 1] 6. 2006 Pamboudian et al. [RATC-3× 3, 1] 7. 2008 Ho et al. [PS – 1× 4, 1] Euclidean distance ≤ 2. 6

國立中興大學 資訊科學與 程學系 Imperceptibility Evaluation (2) 2003 Chen 2002 Tseng[Characteristic [F K W-4× 5, 3] value 4× 4, 1] 7

國立中興大學 資訊科學與 程學系 Imperceptiblity Evaluation (3) 20052008 Tsai Ho. [PWLC-1× 6, 1] et al. [PS-1× 4, 1] 8

國立中興大學 資訊科學與 程學系 Measurement of visual quality PSNR fit for gray or color images ? for binary images ? 2004 Wu et al. [Shuffle-3× 3 least block, 1] 2006 Pamboudian et al. [RATC-3× 3, 1] 2007 Yang et al. [4 Indices- 3× 3, 1] Only fit for 3× 3 block 30 well in gray or color but not fit for binary image PSNR=30 flip 1 bit on image with 1000 bits (32× 32) PSNR=40 flip 1 bit on image with 10000 bits (100× 100) PSNR=50 flip 1 bit on image with 100000 bits (316× 316) 2005 Tsai [PWLC-1× 6, 1] 2008 Ho. et al. [PS-1× 4, 1] Per Pixel Distortion fit for all block pattern ≈0. 5 9

國立中興大學 Capacity(1/2) Block size width≤ 2 資訊科學與 程學系 hiding bits/block size Embedding bits (secret) senary 0 Secret: 001011100011 041532 0. 5 0. 95 Hiding bits Variable length Multi-bits 10 Six states for 2 W 2 B block (2× 2+3× 4)/24≈0. 67

國立中興大學 Capacity(2/2) 資訊科學與 程學系 hiding bits/block size Non-reversible 1. 2002 Tseng et al. [F K W-4× 5, 3]=3/20 =0. 15 2. 2003 Chen et al. [Characteristic value-4× 4, 1]=1/16 ≈0. 06 3. 2004 Wu et al. ≈(≤ 0. 01) [Shuffle-3× 3 least block, 1]=1/? =0. 25 4. 2007 Yang et al. [4 Indices- 3× 3, 1] =1/4 Reversible 5. 2005 Tsai et al. [PWLC – 1× 6, 1]=1/6 ≈ 0. 16 6. 2006 Pamboudian et al. [RATC-3× 3, 1]=1/9 ≈0. 11 7. 2008 Ho et al. [PS – 1× 4, 1]=1/4 ≈0. 25 11 Average ≈0. 14

國立中興大學 資訊科學與 程學系 Reversibility( functional, practical ) Difficult for binary image with less embeddable capacity : not functional • differ available blocks from invalid blocks Math. Formula x 01000 x 11111 Embed 0 Compression x 01000 x 10111 Extract 00 ? : part functional and not practical 2006 Pamboudian [RATC-3× 3, 1] Compressed data secret Record block pattern and location Perceptible Bi-linear Interpolation 12 2005 Tsai [PWLC-1× 6, 1] : not practical Side information 2008 Ho et al. [PS – 1× 4, 1] : functional and practical

國立中興大學 資訊科學與 程學系 Part II- Proposed Schemes n Senary Huffman Compression n n Novel Watermarking Authentication Schemes for Binary Images Based on Dual-Pair Block Pixel Patterns n n Capacity -DPB+NUB(1 B 3 W, 3 W 1 B), Distortion -MPM Hiding Information in Binary Images with Complete Reversibility and High Embedding Capacity n 13 Capacity -Variable length, Distortion -external adjustment (MP) A High Capacity Data Hiding in Two-color Images Based on Block Patterns n n Block pattern DPB(2 B 2 W) selection (2× 2, 4× 1) Bi-linear interpolation with functional and practical target

國立中興大學 資訊科學與 程學系 Senary Huffman Compression-pattern (1) Dual-Pair block senary Embedding bits (secret) senary Secret: 00010111001 0052021 (15/28≈0. 54) 14

國立中興大學 資訊科學與 程學系 Senary Huffman Compression –Confidentiality (2) RND(seed), seed as private-key: select embeddable blocks randomly Decrease distortions Advantages even distribution lower the perceptibility Raise compression rate 15

國立中興大學 資訊科學與 程學系 Senary Huffman Compression-example(3) 16

國立中興大學 資訊科學與 程學系 Novel Watermarking Authentication Schemes for Binary Images Based on Dual-Pair Block Pixel Patterns – Pattern Dual-Pair block (DPB) senary Embedding bits (secret) senary Secret: 001011100011 041532 (16/24≈0. 67) 17

國立中興大學 資訊科學與 程學系 External Adjustment(MP_EA)(1/2)-DPB Distortion Table Adjust mapping relationship between embedding bits and senary Embedding Embedded Alterations (bits) 0 4 2 2 20 2 22 0 2 18 0 15 2 22 2 0 2 22 2 2 4 04 2 4 4 2

國立中興大學 資訊科學與 程學系 External Adjustment(MP_EA) - Illustration (2/2) Embedding frequency table DPB distortion table MP_EA table Original distortions : 4106+524+500+246+522=6402 The least distortions =1790+524+528+472+354+844=4512 ratemp_reduced = (6402 -4512)/6402 100% = 29. 5% MP-table: 5, 1, 3, 4, 0(, 2) 19

國立中興大學 資訊科學與 程學系 MP_EA –Example (1/2) image Stego Host senary 215144532 Embedding Secret (25 bits): 011 010 100 11 011 100 010 101 00 senary 20 213523140 Embedding RND(K) 1, 2, 4, 3, 8, 9, 7, 6, 5 Per pixel distortion = 14/25 >0. 5 (? ) 215340132 Per pixel distortion = 6/25 <0. 5

國立中興大學 資訊科學與 程學系 MP_EA –Example (2/2) Stego Host 245244542 Embedding Secret (24 bits): 100 00 00 100 010 010 303003111 RND(K) ? Per pixel distortion = 18/24 >0. 5 (? ) MP_EA 1, 2, 8, 3, 4, 5, 7, 9, 6 301300113 21 245244552 3 2 0 4 1 5 Per pixel distortion = 2/24 <0. 5

國立中興大學 Capacity/image (‰) Experiment Results (2/2) AVG: 32. 7/((2. 9+6. 2+1+18. 9)/4)= 4. 51 2 nd : 32. 7/18. 9= 1. 73 24 資訊科學與 程學系 AVG: 0. 485/((0. 86+0. 505+0. 493)/4)= 0. 824 2 nd : 0. 485/0. 493= 0. 98

國立中興大學 資訊科學與 程學系 A High Capacity Data Hiding in Two-color Images Based on Block Patterns - Distortion 1 -1 mapping Embedding 0 Embedded Alterations (bits) 1 2 3 4 5 6 7 1 1 1. Internal adjustment 9 A B C D Average Alteration: 2 bits (per pixel distortion=2/4=0. 5) 6 3 8 3 2 2 0 1 3 3 1 2 2 2 -1 mapping, select least distortion block 4 0 5 6 2 0 Embedding 2. External adjustment Embedded 25 Alterations (bits) 1 1 2 6 1 1 3 4 Average Alteration: 1. 1 bits (per pixel distortion=1. 1/4=0. 29) 1 2

國立中興大學 資訊科學與 程學系 A High Capacity Data Hiding in Two-color Images Based on Block Patterns (3 B 1 W, 1 B 3 W) (2 B 2 W) Embedding bits (secret) Septenary Block capacity = (3× 6+2× 1)/7=20/7=2. 857 bits/block (or 0. 71) (Not only capacity ↑compared to DPB 2. 67 bits/block( or 0. 67), but also distortions↓guranteed at most 2 bit-alteration/block. (or 0. 29<0. 5) 26

國立中興大學 資訊科學與 程學系 Maximum pair matching method Least-distortion Table host 27 Alteration between (embedded s 1, embedding s 2): 0: s 1= s 2 s 1≠ s 2 1: (2 B 2 W, 3 B 1 W or 1 B 3 W), (3 B 1 W or 1 B 3 W, 2 B 2 W) 2: (3 B 1 W or 1 B 3 W, 3 B 1 W or 1 B 3 W), (2 B 2 W, 2 B 2 W)

國立中興大學 資訊科學與 程學系 Maximum Pair Matching (MPM) - internal adjustment Image 0 1 Septenary 2 3 4 5 6 3560462 Embedding Secret(20 bits): Septenary 2542436 010 101 100 010 100 011 11 Per pixel distortion: 8/20=0. 4 RND(k) (Alteration: 2+0+2+2+0+1+1) 28 3524462 Per pixel distortion: 2/20=0. 1 6, 2, 4, 3, 5, 7, 1 (Alteration: 0+0+1+1+0+0+0)

國立中興大學 資訊科學與 程學系 Maximum Pair Matching (MPM) - external adjustment Least-distortion table MPM frequency table Original distortions : 808+769+948+1036+579+640+872=5652 808=0× 13+2×(77+71+40)+1×(162+118+152) 680=2×(13+71+40)+0× 77+1×(162+118+152) 872=1×(64+156+137+159)+2×(142+36)+0× 9 9 The least distortion : 680+567+998+1036+617+640+786=5324 Rate. Reducing = 6. 16% 29

國立中興大學 資訊科學與 程學系 Experiment Results AVG: 123/((70+10+1. 4+25)/4)=4. 62 (4. 51 4. 62) 2 nd: 123/70=1. 76 (1. 73 1. 76) 30 AVG: 0. 36/0. 50=0. 72 (0. 824 0. 72) 2 nd: 0. 36/0. 49=0. 73 (0. 98 0. 73)

國立中興大學 資訊科學與 程學系 Hiding Information in Binary Images with Complete Reversibility and High Embedding Capacity 31

國立中興大學 資訊科學與 程學系 Scheme 2(2/2)- embedding criteria (1) The four adjacent pixel blocks, number 1 to 4, are not selected for data hiding, (2) the upper left pixels of block 1 and 2 are not both equal to “ 0”. 32

國立中興大學 資訊科學與 程學系 Example of data embedding (a) interpolated cover image 33 (b) the septenary values to be hidden (c) 101100001011101010 (18 bits) data hiding result

國立中興大學 資訊科學與 程學系 Experimental Results 14% 20% 25% 70 sec ≤ 1 sec 34

國立中興大學 資訊科學與 程學系 Experiment Results (2/2) Distortion : Capacity≈1: 2. 8(Proposed) ≈1: 2(Nonreversible)≈1: 1(Reversible) Per pixel distortion Average : 0. 36(proposed), 0. 5(non-reversible), >1(reversible) 35

國立中興大學 資訊科學與 程學系 Conclusions High information capacity, better visual quality, full reversibility, high confidentiality, and efficiency. l Capacity Ø (hiding bits/block size: proposed 0. 72, others Avg. 0. 14) raise 4. 62 times. l Visual quality Ø Imperceptibility-artifact, Euclidean dist. ≤ 2 2× 2 block pattern DPB, NUB Ø Distortion- (per pixel distortion: proposed 0. 29, others Avg. 0. 5) reduce 0. 72 times. l Reversibility Ø SHC outperforms prior arts Ø Bi-linear interpolation method achieves functional and practical target. l Confidentiality-seed as private-key, block retrieved by PRN(seed) Ø decrease distortions Ø even distribution l Efficiency- algorithmic simplicity 36 36

國立中興大學 資訊科學與 程學系 Future works n n Extend the proposed information hiding methods to other image formats n gray-scale n color Apply the proposed information hiding schemes to other applications 37 n Tamper detection and recovery n Visual secret sharing 37

國立中興大學 資訊科學與 程學系 Thanks for your attention 38