| 摘要 |
<p>HETEROGENEOUS FEATURE FILTERING 5OF THE DISCLOSURE A score fusion CBIR system may generate a large number of feature descriptors which are not effective for image retrieval. Storing ineffective feature descriptors increases storage requirements and also leads to long query retrieval times. By discarding insignificant feature 10 descriptors, storage requirements are reduced and retrieval times are shortened whilst maintaining accurate image matching performance. Feature descriptor filtering may be achieved through the application of a feature descriptor classifier. The classifier may be trained on matched reference and query images. Feature Feature extractor 1 extractor 2 _ 130 125 135 Feature Feature descriptor 1 descriptor 2 database database 145 140_ 170 Feature Feature matcher 1 matcher 2 mage score 120 fusion Feature Feature 160 extractor 1 extractor 2 Fig. 1</p> |
