1 / 28

Detection of Image Alterations Using Semi-fragile Watermarks

Detection of Image Alterations Using Semi-fragile Watermarks. Eugene T. Lin † , Christine I. Podilchuk ‡ and Edward J. Delp †. † Purdue University School of Electrical and Computer Engineering Video and Image Processing Laboratory ( VIPER) West Lafayette, Indiana

Télécharger la présentation

Detection of Image Alterations Using Semi-fragile Watermarks

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Detection of Image Alterations Using Semi-fragile Watermarks Eugene T. Lin†, Christine I. Podilchuk‡ and Edward J. Delp† †Purdue University School of Electrical and Computer Engineering Video and Image Processing Laboratory (VIPER) West Lafayette, Indiana ‡Bell Laboratories, Lucent Technologies Murray Hill, New Jersey

  2. Overview • Introduction • Image authentication • Fragile watermarks • Robust watermarks • Semi-fragile watermarks • Description of proposed technique • Results • Conclusion

  3. Image Authentication • Identify the source of an image • Determine if the image has been altered • If so, locate regions where alterations have occurred • Authentication watermark • watermark is imperceptible under normal observation • allows user to determine if image has been altered after mark embedding

  4. Fragile Watermarks • Watermark is rendered undetectable after slightest modifications to marked content • Typically able to localize alterations with high degree of precision • Sensitivity achieved through use of hash functions • Problem: if lossy compression is applied to marked image, mark is destroyed even though compressed image remains perceptually similar

  5. Robust Watermarks • Resists removal attempts • Examines large regions of image, limited localization of alterations • Robustness typically achieved through spread-spectrum techniques • Problem: robust watermark may remain even after alterations that change the visual message conveyed by the image

  6. Semi-Fragile Watermarks • Able to detect and localize significant “information altering” transformations (feature replacement) • Able to tolerate some degree of “information preserving” transformations (lossy compression) • Suitable in authentication applications where legitimate use includes lossy compression or other image adjustment by users

  7. Semi-Fragile Watermarks • Challenges for fragile watermark  semi-fragile watermark: • LSB plane embedding not tolerant to compression • Cryptographic hash functions too sensitive • Challenges for robust watermark  semi-fragile watermark: • Reduce region size used in mark detection but retain enough SNR to achieve reliable detection • Boundary effects

  8. Description of Proposed Technique • Watermark construction • DCT construction, spatial embedding • Watermark detection • Based on differences of adjacent pixel values • Most natural images contain large regions of relatively smooth features

  9. Watermark Construction DCT Watermark Generation

  10. DCT watermark Generation IDCT W X Marked Image Original Image + Y=X+W Watermark Construction • After watermark is constructed in DCT domain, it is transformed to spatial domain and embedded

  11. Watermark Detection • Independent detection performed on each block, for localizing altered blocks • Define two operators:

  12. Example of Differential Operators

  13. Watermark Detection • Tb = Block of image being tested • Wb = Corresponding block of watermark image • Detector uses both row and column differences:

  14. Block Test Statistic • Tb* and Wb* are correlated to compute block test statistic b: b T: Block is likely authentic b < T: Block is likely altered.

  15. Results - Gradient Original “Gradient” Altered “Gradient” Total Blocks: 682, Altered:300 (44%) Detector Block size:16x16, embedding =5.0

  16. Results - Gradient

  17. Results - Gradient

  18. Results - Sign Original “Sign” Altered “Sign” Total Blocks: 1536, Altered:77 (5%) Detector Block size:16x16, embedding =5.0

  19. Results - Sign

  20. Results - Sign

  21. Results - Money Original “Money” Altered “Money” Total Blocks: 570, Altered:143 (25%) Detector Block size:16x16, embedding =5.0

  22. Results - Money

  23. Results - Money

  24. Results - Girls  Original “Girls” Altered “Girls”  Total Blocks: 5704, Altered:951 (17%) Detector Block size:16x16, embedding =5.0

  25. Results - Girls

  26. Results - Girls

  27. Detection Performance Embed: =5.0 Detection: T=0.1 blocksize=16x16 JPEG-60 bitrate=0.90 bpp 93% correct detection 4% false positive 17% misses

  28. Conclusions • A semi-fragile watermarking technique was proposed which classifies about 70%of blocks correctly for moderate JPEG compression, 90% for light JPEG compression • Detector has problems with edges and textures • Future work: • Integrate a visual model to embed mark at higher strengths in textured areas

More Related