1 / 10

Andrew N. Stein∗ Thomas S. Stepleton Martial Hebert

Towards Unsupervised Whole-Object Segmentation: Combining Automated Matting with Boundary Detection. Andrew N. Stein∗ Thomas S. Stepleton Martial Hebert The Robotics Institute, Carnegie Mellon University Reporter: Hsieh Chia-Hao Date: 2009/09/28. Introduction.

marlie
Download Presentation

Andrew N. Stein∗ Thomas S. Stepleton Martial Hebert

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. Towards Unsupervised Whole-Object Segmentation:Combining Automated Matting with Boundary Detection Andrew N. Stein∗ Thomas S. Stepleton Martial Hebert The Robotics Institute, Carnegie Mellon University Reporter: Hsieh Chia-Hao Date: 2009/09/28

  2. Introduction • Towards Unsupervised Whole-Object Segmentation: Combining Automated Matting with Boundary Detection

  3. Outline • Segmentation “Hints” via Multiple Mattes • α-Matting • Multiple Mattes  Affinities • Detecting Boundary Fragments • Image Segmentation by Ncuts • Evaluating Object Segmentations • Experiments

  4. α-Matting Multiple Mattes • α-Matting • Multiple Mattes  Affinities Pixel OR super-pixels

  5. Detecting Boundary Fragments • Incorporating local, instantaneous motion estimates when short video clips are available • Over-segments a scene into a few hundred “super-pixels” using a watershed-based approach • Learned classifiers and inference on a graphical model  boundary probabilities (weight for W)

  6. Image Segmentation by Ncuts • Given pair-wise affinity matrix A • Use spectral clustering according to the Normalized Cut Criterion • Produce an image segmentation with K segments • K is variable

  7. Evaluating Object Segmentations • Consistency • Efficiency R = {A, B, C, . . . } ⊆S is a subset of segments from a given (over-)segmentation S

  8. Experiments

  9. Experiments bottom-up perspective

  10. Discussion • Promising results for use in subsequent work on unsupervised object discovery or scene understanding

More Related