| Article ID | Journal | Published Year | Pages | File Type |
|---|---|---|---|---|
| 408908 | Neurocomputing | 2008 | 12 Pages |
Abstract
We introduce an image recognition system that does not require availability of a complete training data. The system consists of a constrained K-Means clustering algorithm and an image recognition neural network. For finding similarity between images we use a novel image similarity measure and introduce a new image cluster validity measure to determine the most probable number of clusters. Extensive testing on several image datasets indicates good performance of the system.
Keywords
Related Topics
Physical Sciences and Engineering
Computer Science
Artificial Intelligence
Authors
Chia-Yu Yen, Krzysztof J. Cios,