Title Page
Abstract
Contents
Chapter 1. Introduction 15
Chapter 2. Related Work 22
2.1. Prediction-Error Method 22
2.1.1. Prediction-Error Expansion Mechanism 23
2.1.2. Prediction Algorithm 24
2.1.3. Coding Procedure 25
2.2. Pattern Substitution Method 27
Chapter 3. Proposed Method 33
3.1. PFR method 33
3.2. Improved Overlapping Pattern Substitution Method 37
3.2.1. Three Counterexamples 37
3.2.2. Changeability 38
3.2.3. Embeddability 42
3.2.4. Run length Coding for Location Map 44
3.2.5. Encoding and Decoding 45
3.3. Adaptive Pattern Substitution Method 45
3.3.1. Adaptive Embedding 46
3.3.2. Two Embedding Modes 47
3.3.3. Three-Round Embedding 49
Chapter 4. Experimental Results 52
4.1. Analysis of PFR Method 53
4.2. Analysis of the improved overlapping PS method 53
4.3. Analysis of the adaptive PS method 57
4.3.1. Adaptive Embedding 57
4.3.2. Three-round Embedding 58
4.3.3. Comparison with the previous works 58
4.4. Conclusion of Experiments 58
Chapter 5. Applications 63
5.1. Reversible Annotation 63
5.2. Tamper Detection 64
5.2.1. Tamper Detection 64
5.2.2. Tamper Localization 66
5.2.3. Security Analysis 68
Chapter 6. Conclusions 74
6.1. Summary of Contributions 74
6.2. Future Works 75
6.3. List of Publications 75
Bibliography 77
Table 2.1: Pattern and its suitable patterns for substitution. 31
Table 4.1: Pattern's number information. 61
Table 4.2: Comparisons of the LM size. 62
Table 4.3: Standard deviation based on the PSNR values with different... 62
Table 4.4: The size of location map with different size of the SD embedded 62
Table 5.1: Detection probability with different tampering ratios and block... 73
Fig 2.1: The context of the pixel X 24
Fig 2.2: An image matrix and its difference matrix. 28
Fig 2.3: A pattern P0001 and its suitable patterns for substitution.(이미지참조) 30
Fig 3.1: Three counterexamples to Ho et al.'s rules. 37
Fig 3.2: An illustration of class map where PM is 0001 and PF is 0111. 39
Fig 3.3: An example of pattern that is not changeable when PM is 0001 and PF is 0111. 41
Fig 3.4: A Changeable but not embeddable pattern when PM is 1001 and PF is 1111, and... 43
Fig 3.5: An illustration of the context window. 47
Fig 3.6: The three-round embedding mechanism. 51
Fig 4.1: The test images. 52
Fig 4.2: The capacity-distortion performance of the proposed method... 54
Fig 4.3: Comparison result on the number of flippable pixels. 56
Fig 4.4: The capacity-distortion performance of the proposed method... 56
Fig 4.5: The capacity-distortion performance of the proposed method... 59
Fig 5.1: Invisible annotation with free distortion: (a) original image; (b)... 64
Fig 5.2: Authentication results: (a) original image; (b) stego-image with... 65
Fig 5.3: Authentication results: (a) original image; (b) stego-image with... 69