1 / 23

Image Categorization Dog or Cat

Image Categorization Dog or Cat. Grace Akpan Hehe Feng Junci Wang Mehran Javanmardi Yuanyuan Gao. Kaggle http ://www.kaggle.com/c/dogs-vs-cats Microsoft Asirra CAPTCHA Caltech (object recognition). Introduction . Detecting whether image contains a dog or cat. Purpose.

olinda
Download Presentation

Image Categorization Dog or Cat

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. Image CategorizationDog or Cat Grace Akpan Hehe Feng Junci Wang MehranJavanmardi YuanyuanGao

  2. Kaggle http://www.kaggle.com/c/dogs-vs-cats • Microsoft Asirra CAPTCHA • Caltech (object recognition) Introduction

  3. Detecting whether image contains a dog or cat Purpose

  4. Why we are interested in this problem? • Classification Method used in the previous articles • Challenge: Feature Extraction • Feature Extraction Methods we determine to use: • Scale Invariant Feature Transform (SIFT) • K – means Clustering • Bag of Visual Words • Image Tiling Problem Definition

  5. Dense key points • Sift descriptors • K – means Clustering • Bag of words: Histogram • Tiling (creating sub-images) • Spatial histograms for each sub-image Procedure

  6. 10 Dense Key Points 10

  7. Sift Descriptors • Noise Robust • Scale Invariant • Rotation Invariant • Viewpoint Invariant • Illumination Invariant

  8. K – means Clustering Bag of words

  9. Tiling

  10. Spatial histograms Bag of words: 4000*4 + 4000 = 20000 # of clusters *4 + # of clusters

  11. Whole Procedure

  12. Given: • 1 : training 4000 images containing cat • 0: training 4000 images containing dog • Classify: Train/Apply the Model ?

  13. Accuracy (1000 images of dog and 1000 images of cat) Results/Conclusion

  14. The size of the pictures • The orientation of the pictures • Irrelevant pictures • Hyper-parameters • Memory problem • Long processing time Discussions

  15. Test with some refined data (only face of cat/dog): • Add face detection Improvement/ Future Work Training set: 660 Testing set: 110 Clusters: 50 Accuracy :76.4% Verifying accuracy (test on training set): 99.54%

  16. Open source library provided by VLFeat.org http://www.vlfeat.org/ • Suggest strategy provided by Prof. Andrea Vedaldi and Andrew Zisserman research group from University of Oxford http://www.robots.ox.ac.uk/~vgg/share/practical-image- classification.htm • Parkhi, Omkar M., et al. "The truth about cats and dogs." Computer Vision (ICCV), 2011 IEEE International Conference on. IEEE, 2011. • Belokurov, V., et al. "Cats and dogs, hair and a hero: a quintet of new Milky Way companions." The Astrophysical Journal 654.2 (2007): 897. • Golle, Philippe. "Machine learning attacks against the AsirraCAPTCHA."Proceedings of the 15th ACM conference on Computer and communications security. ACM, 2008. External Sources/ References

  17. Dr. Jurvan den Berg • Dustin Webb / Brig Bagley / Liang He • Each Team member Yuanyuan Acknowledgement Hehe Mehran Junci Grace

  18. Questions?

  19. SIFT details

More Related