Not Logged In

PapersDB

Learning Policies for Efficiently Identifying Objects of Many Classes

• Ramana Isukapalli
• Ahmed Elgammal, Rutgers University
• Russ Greiner, Dept of Computing Science; PI of AICML

Viola and Jones (VJ) cascade classification methods have proven to be very successfulin detecting objects belonging to a single class --- eg faces. This paper addresses the more challenging ``many class detection'' problem:detecting and identifying objects that belong to any of a set of classes. We use a set of learned weights (corresponding to the parameters of a set of binary linear separators)to identify these objects.We show that objects within many real-world classestend to form clusters in this induced ``classifier space''. As the result of a sequence of classifiers can suggest a possible label foreach object, we formulate this task as a Markov Decision Process. Our system first uses a ``decision tree classifier, a policy produced using dynamic programming)to specify when to apply which classifierto produce a possible class label for each sub-image W of a test image.This corresponds to a leaf of the decision tree.It then uses a cascade of classifiers, specific to this leaf to confirm that W is an instance of the proposed class.We present empirical evidence to verify that our ideas work effectively:showing that our system is essentially as accurate as running a set of cascadeclassifiers (one for each class of objects), but is much faster than that approach.

Citation

BibTeX

@incollection{Isukapalli+al:ICPR06,
  author = {Ramana Isukapalli and Ahmed Elgammal and Russ Greiner},
  title = {Learning Policies for Efficiently Identifying Objects of Many
    Classes},
  booktitle = {International Conference on Pattern Recognition (ICPR)},
  year = 2006,
}

Last Updated: April 23, 2007
Submitted by Nelson Loyola

• Home
• Search
  • Advanced Search
• Publications
  • View Publication Entries
• Authors
  • View Authors
• User
  • Login or Register