1. Object Class Recognition by Unsupervised Scale-Invariant Learning R. Fergus, P. Perona, and A. Zisserman Presented By Jeff

2. Goal: Enable Computers to Recognize Different Categories of Objects in Images.

4. Components Model –Generative Probabilistic Model –Location, Scale, and Appearance Learning –Estimate Parameters Via EM Recognition –Evaluate Image Using Model and Threshold

5. Model: Constellation Of Parts Fischler & Elschlager, 1973  Yuille, ‘ 91  Brunelli & Poggio, ‘ 93  Lades, v.d. Malsburg et al. ‘ 93  Cootes, Lanitis, Taylor et al. ‘ 95  Amit & Geman, ‘ 95, ‘ 99  Perona et al. ‘ 95, ‘ 96, ’ 98, ‘ 00

6. Parts Selected by Interest Operator • Kadir and Brady's Interest Operator. • Finds Maxima in Entropy Over Scale and Location

7. Representation of Appearance 11x11 patch c1c1 c2c2 Normalize Projection onto PCA basis c 15

8. Generative Probabilistic Model Start with Recognition:

9. Appearance Gaussian Part Appearance PDFGaussian Appearance PDF

10. Shape Gaussian Shape PDFUniform Shape PDF

11. Scale Prob. of detection Gaussian Relative Scale PDF Log(scale) 0.80.750.9 Poission PDF On # Detections Uniform Relative Scale PDF Log(scale)

12. Occlusion and Part Statistics

13. Learning Train Model Parameters Using EM: Optimize Parameters Optimize Assignments Repeat Until Convergence

14. Recognition Make This: Greater Than Threshold

15. RESULTS

16. Motorbikes Equal error rate: 7.5%

17. Background Images

18. Frontal faces Equal error rate: 4.6%

19. Airplanes Equal error rate: 9.8%

20. Scale-Invariant Cats Equal error rate: 10.0%

21. Scale-Invariant cars Equal error rate: 9.7%

22. Robustness of Algorithm

23. ROC equal error rates Pre-Scaled Data (Identical Settings): Scale-Invariant Learning and Recognition:

24. Scale-Invariant Cars

25. The End