Interactive multimedia system for distance learning of higher education
Edutainment'06 Proceedings of the First international conference on Technologies for E-Learning and Digital Entertainment
International Journal of Multimedia Data Engineering & Management
| Hi-index | 0.00 |
Feature selection for video categorization isimpractical with existing techniques. In this paper wepresent a novel algorithm to select a very small subset ofimage features. We reduce the cardinality of the input databy sorting the individual features by their effectiveness incategorization, and then merging pairwise these featuresinto feature sets of cardinality two. Repeating this sort-mergeprocess several times results in the learning of asmall-cardinality, efficient, but highly accurate feature set.The cost of this wrapper method for learning the featureset, approximately O(F logF) where F is the number ofincoming features, is very reasonable, particularly whencompared with the impracticality of applying much highercost current filter or wrapper learning models to themassive data of this domain. We provide empiricalvalidation of this method, comparing it to both randomand hand-selected feature sets of comparable smallcardinality.