Graph-based classification of multiple observation sets
Pattern Recognition
Improving object classification in far-field video
CVPR'04 Proceedings of the 2004 IEEE computer society conference on Computer vision and pattern recognition
Learning semantic scene models by trajectory analysis
ECCV'06 Proceedings of the 9th European conference on Computer Vision - Volume Part III
| Hi-index | 0.00 |
This paper discusses building complex classifiers from a singlelabeled example and vast number of unlabeled observation sets, eachderived from observation of a single processor object. When datacan be measured by observation, it is often plentiful and it isoften possible to make more than one observation of the state of aprocess or object. This paper discusses how to exploit thevariability across such sets of observations of the same object toestimate class labels for unlabeled examples given a minimal numberof labeled examples. In contrast to similar semi-supervisedclassification procedures that define the likelihood that twoobservations share a label as a function of the embedded distancebetween the two observations, this method uses the Naive Bayesestimate of how often the two observations did result from the sameobserved process. Exploiting this additional source of informationin an iterative estimation procedure can generalize complexclassification models from single labeled observations. Someexamples involving classification of tracked objects in alow-dimensional feature space given thousands of unlabeledobservation sets are used to illustrate the effectiveness of thismethod.