1. A Comparison Between Bayesian Networks and Generalized Linear Models in the Indoor/Outdoor Scene Classification Problem

2. Overview Introduce Scene Classification Problems Motivation for Scene Classification Kodak's JBJL Database and Features Bayesian Networks  Brief Overview (description, inference, structure learning)‏  Classification Results GLM  Briefer Overview  Classification Results Comparison and Conclusion

3. Problem Statement: Given a set of consumer digital images, can we use a statistical model to distinguish between indoor images and outdoor images?

4. Motivation Kodak  Increase visual appeal by processing based on classification Object Recognition  Provide context information which may give clues to scale, location, identity, etc.

5. Procedure Establish ground-truth for all images Perform feature extraction and confidence/probability mapping for features Divide images into training and testing set Use test images to train a model to predict ground-truth Use the model to predict ground truth for the test set Evaluate performance

6. Kodak JBJL Consumer image database 615 indoor and 693 outdoor images Some images are difficult for HSV to determine whether it is indoor or outdoor Some images have indoor and outdoor parts

7. Features and Probability Mapping “Low-level” Features  Ohta-space color histogram (color information)‏  MSAR model (texture information)‏ “Mid-level” Features  Grass classifier  Sky classifier K-NN Used to Extract Probs from Features  Quantized to nearest 10% (11 states for Mid-level, 3 states for Low-level)‏

8. Feature Probs and Classes

9. Stat. Model 1: Bayesian Network Graphical Model Variables are represented by vertices of a graph Conditional relationships are represented by directed edges Conditional Probability table associated with each vertex  Quantifies vertex relationships  Facilitates automated inference

11. Exact Inference Model Joint Probability Inference

12. Structure Learning Search Space Space BNs  Variable-State Combination (#States per Node) x (#Nodes)‏ 2178 possible  Structures Limited to DAGs 29281

13. Scoring Metric Score a structure based on how well the data models the data We do have an expression estimate the data given the structure Unfortunately, the data probability is difficult to estimate

14. The Bayes Dirichlet Likelihood Equivalent Can compare structures 2 at a time What is the prior on structure?  Assume all structures are equally likely  Use #edges to penalize complex networks

15. Challenges Not all structures can be considered if there is only a small amount of data.  Context dilution  Can't consider cases where CPT cannot be filled in Finding an optimal structure is NP hard

16. BDe Structure For I/O Classification Greedy algorithm with BDe scoring Naïve Bayes Model!

17. Result Compared to Previous Previous Results Our Results

18. Misclassified:Inferred Outdoor

19. Misclassified: Inferred Indoor

20. Generalized Linear Model Outdoor and Indoor can be thought of a binary output Logit kernel

21. Likelihood for GLM Newton-Raphson  Get estimates of mean and variance (1 st and 2 nd derivative)‏  Find optimal based on estimates (Taylor Expansion)‏  Iterate Generally, this quickly converges to the optimal solution

23. Side by Side Comparison

24. Misclassified: Predicted Outdoor

25. Misclassified: Predicted Indoor

26. Conclusion The newer Bayesian Network model may perform classification slightly better than GLM  BN is more computationally intensive  Unclear if there is in fact a difference  Both models have difficulty with the same images Better to introduce new data than to use a new model  New model give (at most) marginal improvement

27. References Heckerman, D. A Tutorial on Learning with Bayesian Networks. In Learning in Graphical Models, M. Jordan, ed.. MIT Press, Cambridge, MA, 1999. Murphy, K. A Brief Introduction to Graphical Models and Bayesian Networks, http://www.cs.ubc.ca/~murphyk/Bayes/bnintro.h tml(viewed 4/1/08)‏ http://www.cs.ubc.ca/~murphyk/Bayes/bnintro.h tml Lehmann, E.L. and Casella G. Theory of Point Estimation (2 nd edition)‏ Weisberg, S. Applied Linear Regression (3 rd Edition)‏

28. Data Given Model Prob