1 / 27

Spatial Pyramid Co-occurrence for Image Classification

Spatial Pyramid Co-occurrence for Image Classification. Presenter : Han-Mu Park. Spatial Pyramid Co-occurrence for Image Classification, ICCV 2011. Contents. Introduction Coding methods Proposed method Experimental results Conclusion References.

mikko
Télécharger la présentation

Spatial Pyramid Co-occurrence for Image Classification

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. Spatial Pyramid Co-occurrence for Image Classification Presenter : Han-Mu Park

  2. Spatial Pyramid Co-occurrence for Image Classification, ICCV 2011 Contents • Introduction • Coding methods • Proposed method • Experimental results • Conclusion • References

  3. Spatial Pyramid Co-occurrence for Image Classification, ICCV 2011 Introduction • Bag-of-Words (BoW) model • An image is represented as a collection of visual words. • Generally, to represent the collection, histogram of words form is used.

  4. Spatial Pyramid Co-occurrence for Image Classification, ICCV 2011 Introduction • Spatial Pyramid Matching (SPM) • Original SPM partitions an image into a sequence of spatial grids at resolutions . • The grid at level has cells along each dimension for a total of cells. • Some variations used different shaped partitions. • Overlapped partitions • Vertically divided parts (ex) 3x1, 3x2) Example of a tree-level SPM Example of overlapped partitions [J.WU2012]

  5. Spatial Pyramid Co-occurrence for Image Classification, ICCV 2011 Introduction • Testing images • SPM can extract the characteristics of “car”.

  6. Spatial Pyramid Co-occurrence for Image Classification, ICCV 2011 Introduction • Testing images • SPM can extract the characteristics of “car”…?

  7. Spatial Pyramid Co-occurrence for Image Classification, ICCV 2011 Introduction • Testing images • SPM can extract the characteristics of “car”…?

  8. Spatial Pyramid Co-occurrence for Image Classification, ICCV 2011 Introduction • Correlation between codewords • Correlogram extracts frequent pattern information from the co-occurrence of codewordsin the local region. • Correlation between codewords is extracted from correlogram by mining the primitive patterns. Example of local regions [S.Savarese2006] (a) Image (b) codeword image (c) correlogram V(1,2) [S.Savarese2006]

  9. Spatial Pyramid Co-occurrence for Image Classification, ICCV 2011 Introduction • Motivation • Both of spatial characteristics, absolute and relative, have to be considered. • Various local spatial arrangements should be handled. • The method can be easily combined with conventional framework.

  10. Spatial Pyramid Co-occurrence for Image Classification, ICCV 2011 Proposed method • Bag-of-Visual-Words (BOVW) representation • The non-spatial BOVW representation simply records the visual word occurrences in an image. • It is typically represented as a histogram • : # of occurrences of visual word • BOVW kernel (intersection kernel)

  11. Spatial Pyramid Co-occurrence for Image Classification, ICCV 2011 Proposed method • Spatial Pyramid representation • The spatial pyramid representation partitions an image into a sequence of spatial grids at resolutions . • Such that the grid at level has cells. • A Spatial Pyramid Match Kernel (SPMK)

  12. Spatial Pyramid Co-occurrence for Image Classification, ICCV 2011 Proposed method • Spatial Co-occurrence representation • A binary spatial predicate where is defined. • The Visual Word Co-occurrence Matrix (VWCM) is defined as a count of the number of times two visual words satisfy the spatial predicate. • Spatial Co-occurrence Kernel (SCK)

  13. Spatial Pyramid Co-occurrence for Image Classification, ICCV 2011 Proposed method • Spatial predicates • Proximity • Orientation

  14. Spatial Pyramid Co-occurrence for Image Classification, ICCV 2011 Proposed method • Spatial predicates

  15. Spatial Pyramid Co-occurrence for Image Classification, ICCV 2011 Proposed method • Combining multiple spatial predicates • Multiple binary spatial predicates can be easily combined. • The combined SCK is simply computed as the sum of the individual SCKs.

  16. Spatial Pyramid Co-occurrence for Image Classification, ICCV 2011 Proposed method • Spatial Pyramid Co-occurrence representation • A spatial predicate is computed for each cell • Spatial Pyramid Co-occurrence Kernel (SPCK) , where

  17. Spatial Pyramid Co-occurrence for Image Classification, ICCV 2011 Proposed method • Extended SPCK • The extended SPCK merges SPCK and non-spatial BOVW or SPMK. • Extended SPCK+ (SPCK & non-spatial BOVW) • Extended SPCK++ (SPCK & SPMK)

  18. Spatial Pyramid Co-occurrence for Image Classification, ICCV 2011 Experimental results • Evaluation data sets • Land-Use data set • Aerial orthoimagery • 256 x 256 pixel • 21 classes • Agricultural, airplane, baseball diamond, beach, …

  19. Spatial Pyramid Co-occurrence for Image Classification, ICCV 2011 Experimental results • Evaluation data sets • GRAZ-01 data set • High intra-class variation • 640 x 480 pixel • 3 classes • Bike, Person, Background

  20. Spatial Pyramid Co-occurrence for Image Classification, ICCV 2011 Experimental results • Evaluation data sets • 15 scene data set • Images in the same class have similar composition. • 300 x 300 pixel • 15 classes • Bedroom, Kitchen, coast, city, forest, …

  21. Spatial Pyramid Co-occurrence for Image Classification, ICCV 2011 Experimental results • Land-Use data set

  22. Spatial Pyramid Co-occurrence for Image Classification, ICCV 2011 Experimental results • Land-Use data set

  23. Spatial Pyramid Co-occurrence for Image Classification, ICCV 2011 Experimental results • Land-Use data set

  24. Spatial Pyramid Co-occurrence for Image Classification, ICCV 2011 Experimental results • Graz-01 data set • 15 Scene data set

  25. Spatial Pyramid Co-occurrence for Image Classification, ICCV 2011 Conclusion • Conclusion • In this paper, Spatial Pyramid Co-occurrence Kernel (SPCK) is proposed. • The proposed method includes absolute and relative spatial information of codewords • The proposed method shows better performance than non-spatial BoVW and SPMK framework

  26. Spatial Pyramid Co-occurrence for Image Classification, ICCV 2011 References [1] Y. Yang and S. Newsam, “Spatial Pyramid Co-occurrence for Image Classification,” ICCV 2011. [2] J. Wang, J. Yang, K. Yu, F. Lv, T. Huang, Y. Gong, “Locality-constrained Linear Coding for Image Classification,” CVPR 2010. [3] J. Wu, M. Rehg, “CENTRIST: A Visual Descriptor for Scene Categorization,” PAMI 2011. [4] S. Savarese, J. Winn, A. Criminisi, “Discriminative Object Class Models of Appearance and Shape by Correlations,” CVPR 2006.

  27. Thank you!

More Related