1 / 28

Chenlei Guo Liming Zhang Image Processing 2010

A Novel Multiresolution Spatiotemporal Saliency Detection Model and Its Applications in Image and Video Compression. Chenlei Guo Liming Zhang Image Processing 2010. Outline. Introduction Phase Spectrum of Quaternion Fourier Transform (PQFT)

dane
Download Presentation

Chenlei Guo Liming Zhang Image Processing 2010

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. A Novel Multiresolution Spatiotemporal Saliency Detection Model and Its Applications in Image and Video Compression ChenleiGuo Liming Zhang Image Processing 2010

  2. Outline • Introduction • Phase Spectrum of Quaternion Fourier Transform(PQFT) • Detect Proto-Objects in the Spatiotemporal Saliency Map • Hierarchical Selectivity (HS) • Experiment Result • Applications in Image and Video Coding • Conclusions and Discussions

  3. Introduction • Most traditional object detectors need training • Graph-based visual saliency detection can be very powerful but it demands a very high computational cost • Most of the models only consider static images

  4. Phase Spectrum of Quaternion Fourier Transform(PQFT) (1/3) • Locations with less periodicity or less homogeneity create ”pop out” proto objects in the reconstruction of the image’s phase spectrum • An early saliency detection model : PFT

  5. Quaternion Representation (2/3) • Define the input image captured at time t as F(t) • r(t), g(t), b(t) are color channels of F(t)

  6. Calculate the Saliency Map By PQFT (3/3) 2-D gaussian filter

  7. Detect Proto-Objects (1/3) • : the object candidate area • The search stops when is satisfied

  8. Alpha (2/3)

  9. Gamma (3/3)

  10. How PQFT Select Visual Resolution • PQFT simulates the human vision system(HVS)

  11. Hierarchical Selectivity • Set hierarchical level

  12. Experiment Results • Video Sequence • Natural Images • Psychological Patterns

  13. Video Sequence (1/3)

  14. Video Sequence (2/3)

  15. Video Sequence (3/3)

  16. Natural Image

  17. Evaluation Method - ROC • True Positive Rate(TPR), False Positive Rate(FPR) • Receiver Operating Characteristic (ROC) • ROC curve = TPR/FPR • ROC area = area beneath ROC curve • The larger ROC area is, the better the prediction power of a saliency map.

  18. Psychological Patterns (1/3)

  19. Psychological Patterns (2/3)

  20. Psychological Patterns (3/3)

  21. Applications in Image and Video Coding • Multiresolution Wavelet Domain Foveation Model (MWDF) • Evaluate the performance of the HS-MWDF model in Image and video compression

  22. Multiresolution Wavelet Domain Foveation Model (MWDF) • JPEG 2000 has included the region-of-interest(RoI) coding in drafts • A better way to find RoI:use Hierarchical Selectivity

  23. Multiresolution Wavelet Domain Foveation Model (MWDF)

  24. The Performance of HS-MWDF in Image Compression • We use HS-MWDF model as a front end before standard compression (JPEG 2000) • Set nfov => we only use the first n OCAs found by PQFT Auto fov => let the program itself decide the number

  25. The Performance of HS-MWDF in Video Compression

  26. Conclusion and Discussion • Extend PFT model to PQFT model • PQFT model is independent of parameters and prior knowledge, and is fast enough to meet real-time requirements • Develop a model called HS-MWDF as a front end before the image/video encoder • Problems: • Can’t deal with closure patterns well • Only considers bottom-up information • Insert the model into the image/video encoders

  27. References

More Related