1 / 23

Discriminative Local Binary Patterns for Human Detection in Personal Album

Discriminative Local Binary Patterns for Human Detection in Personal Album. 20101230. The main contribution of our work is along the first direction, i.e. we focus on building more powerful features for human detection.

owena
Download Presentation

Discriminative Local Binary Patterns for Human Detection in Personal Album

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. Discriminative Local Binary Patterns for Human Detection in Personal Album 20101230

  2. The main contribution of our work is along the first direction, i.e. we focus on building more powerful features for human detection.

  3. original LBP descriptor does not suit the human detecting problem well due to its high complexity and lack of semantic consistency, thus we propose two variants of LBP: Semantic-LBP and Fourier- LBP.

  4. INTRODUCTION

  5. LBP

  6. Uniform LBP • An important special case of LBP is the uniform LBP. A LBP descriptor is called uniform if and only if at most two bitwise transition between 0 and 1 over the circulated binary feature. • For example: • 00000000 (0 transition), 11100011 (2 transitions) are uniform • 01010000 (4 transitions), 01110101 (6 transitions) are non-uniform

  7. Multi-scale LBP

  8. WHY USE LBP?

  9. Compare VS gradient • Two same gradients may correspond to rather different local structures, thus ambiguous.

  10. Compare VS gradient • Secondly, the target object typically appears in a cluttered environment, and the unexpected noises will drastically degrade the performance. Only gradient information is insufficient to judge useful points and outliers. • On the contrary, the concept of ”uniform LBP” provides the possibility to effectively remove outliers.

  11. Compare VS gradient • Thirdly, gradient based methods typically drop color information contained in original images, for it is difficult to define a metric for colors similar to intensity gradient.

  12. TWO WAY

  13. S-LBP

  14. F-LBP • s(k) is real valued color distance between the k-th samples and central pixel • S = {s(k), k = 0 . . . P − 1} • A = {a(u), u = 0. . . P −1}

  15. implement • Some early work of human detection [5] only used subwindows with fixed small size (8×8) • adopt the variable-size strategy, and sampling subwindows in a similar way to [14]: the minimum subwindows units are sampled with 1/K

  16. evaluation • Especially, the learning algorithms there are chosen according to following criterions: • • it should be state-of-the-art. • • suitable for a variety of feature types. • • fewer parameters to be tuned are preferred. •  HOG •  COV

  17. evaluation • (1) Adopt cascade-of-rejectgors+logitBoost [8] as our basic learning algorithm and use a L2 normalized 8×8 histogram descriptor.

  18. evaluation • (2) inspect the classification score for each individual subwindow for HOG, COV and our LBP-based descriptors. • Many sub window ,each descriptive vector can be calculated

  19. evaluation

  20. evaluation

  21. evaluation • Z value is widely used to estimate how discriminative a weak classifier is. • A small Z value indicates ”good” data distribution for the classification task

  22. evaluation

  23. THE END

More Related