Images Images are a set of pixels Each

Images � Images are a set of pixels �Each pixel: � 1 bit: binary image (0 or 1) � 8 bit: gray scale image (0 -255) � 24 bit: color image �Intensity, value, …

Image Type �Compression : �Lossless compression � Reconstructed image: numerically identical to the original image �Lossy compression � Reconstructed image: contains degradation relative to the original �Image file format: �BMP �JPEG �GIF �TIFF �…

Images with lossless compression BMP: 768 KB TIFF: 933 KB

Images with lossy compression GIF: 102 KB JPEG: 40 KB

Embedding in spatial domain �Embedding in value of pixels �Human Visual System (HVS) characteristic: �Insensitivity to high frequency region �Insensitivity to redundant data in image Lena image

LSB-F ﻧﻘﻄﻪ ﺿﻌﻒ 2 3 100 50 LSB Cover image 2 3 75 75 Stego image Embedding with LSB Flipping creates POVs have equal frequencies in stego image Chi square attack 16

Chi square ﻧﺘیﺠﻪ ﺣﻤﻠﻪ Result of chi square attack on a cover Image 17

ﻫیﺴﺘﻮگﺮﺍﻡ ﺗﺼﻮیﺮ ﺑﻌﺪ ﺍﺯ ﺟﺎﺳﺎﺯی Image histogram after LSB Flipping embedding 18

Chi square ﻧﺘیﺠﻪ ﺣﻤﻠﻪ Result of chi square attack on a Image after LSB Flipping embedding 20

ﻫیﺴﺘﻮگﺮﺍﻡ ﺗﺼﻮیﺮ ﺑﻌﺪ ﺍﺯ ﺟﺎﺳﺎﺯی Image histogram after LSB Matching embedding 22

Chi square ﺣﻤﻠﻪ Result of chi square attack on a Image after LSB Matching embedding 24

LSB-M ﺑﻬﺒﻮﺩ LSB-M � ﺑﻬﺒﻮﺩ ﺍﻣﻨیﺖ ﻭ ﻇﺮﻓیﺖ High capacity LSB Matching (H-LSB-M) Adaptive LSB Matching (A-LSB-M) Adaptive high capacity LSB Matching (AH-LSB-M) Revised LSB Matching (R-LSB-M) Generalized LSB Matching (G-LSB-M) Content-Adaptive Steganography via Denoising (CAS-D) Content-Adaptive Steganography via Noise Estimation (CAS-NE) � � � � 26

High Capacity LSB Matching �A method for capacity improvement: �Using of m bit for embedding �Pixel change range: 100 01 10 97 98 11 00 01 10 11 99 100 101 102 103 m=2 27

Adaptive LSB Matching �A method for security improvement: �adaptively choose pixel values no randomly. �Aim: minimizing histogram alterations. 90 → 91 ? 110 100→ → 111? 101? �Example: 105 → 106? �Embedding 1 in 126: 125 or 127? 125 �Embedding 0 in 129: 128 or 130? 128 Pixel intensity Population in cover image Population in Stego image 131 100 130 100 129 100 128 100 127 100 126 100 125 100 120 105 90 100 110 100 90 28

PVD ﻣﺜﺎﻟی ﺍﺯ 1010 Data=10101110010100011010101100 B 1: (25, 50) d= 50 -25=25 B 1’: (24, 50) b=(1010)=10 d'=16+10=26 30

ﻫیﺴﺘﻮگﺮﺍﻡ ﺗﻔﺎﻭﺕ ﺗﺼﺎﻭیﺮ ﻋﺎﺩی Difference histogram of image without embedding 32

ﺭﻭﺵ ﻫﺎی ﺗﻄﺎﺑﻘی Pu Pl Px Side match � ﺭﻭﺵ Two-Side � Three-Side � Four-Side � 35

ﺭﻭﺵ ﻫﺎی ﺣﻮﺯﻩ ﺗﺒﺪیﻞ Embedding in spatial domain Lossy compression Compressed Uncompressed Embedded Data stego image Uncompressed image Embedding Data Compressed stego image = Extracted Data ? Uncompressed stego image Decompress Extracted Data Extraction 37

ﺭﻭﺵ ﻫﺎی ﺣﻮﺯﻩ ﺗﺒﺪیﻞ JPEG � Forward transform (DCT) � Quantization � Lossless compression � DC coefficient AC coefficients DCT Quantization � � Lossless compression 38

ﺭﻭﺵ ﻫﺎی ﺣﻮﺯﻩ ﺗﺒﺪیﻞ DCT + Quantization Data Lossless Compression Uncompressed image DCT Coefficients Compressed image+Data (Stego image) Embedding in Transform Domain 39

43

Chi square attack 44

PVD Embedding �Embedding steps: 1. Zigzag scan and partition image into 2 pixel blocks 2. For each block 1. 2. 3. 4. Calculate block difference Determine the number of bits for embedding Calculate new difference after embedding Calculate pixels value in stego image 45

Results of PVD attack Attack result on Image without embedding 46

Results of PVD attack Attack result on Image with PVD embedding 47

F 3 algorithm �Mechanism: �Decrement the non-zero coefficient’s absolute value only if the LSB does not match. �Zero coefficients are skipped. �Characteristic: �statistical attack (chi-square test) will not be successful �Less capacity and surplus of even coefficients caused by shrinking 48

F 3 Shrinkage �Since the receiver cannot tell between a skipped zero and a zero that was generated due to shrinkage, repetitive embedding is necessary. -4 0 -4 -3 -2 -1 0 1 1 0 -3 -2 -1 1 0 0 2 3 4 11 0 0 1 1 1 2 3 0 4 Shrinkage 49

F 3 algorithm Original F 3 Algorithm JSTEG 50

F 4 algorithm �Main idea: �Mapping negative coefficients to the inverted steganographic value. �Even negative and odd positive coefficients: one. �Even positive and odd negative coefficients: zero. 51

F 4 algorithm �Mechanism: If LSB and message does not match, � Increment negative Coefficients � Decrement positive coefficients �Example: Embed the code “ 0111” Original Coefficient Extracted Bit Message Stego Coefficient 5 (0101) 1 0 4 (0100) -2 (1110) 1 1 -2 (1110) 3 (0011) 1 1 3 (0011) -1 (1111) 0 1 0(0000) -3 (1101) 0 1 -2 (1110) 52