The query by image content (QBIC) system
SIGMOD '95 Proceedings of the 1995 ACM SIGMOD international conference on Management of data
Content-Based Image Retrieval at the End of the Early Years
IEEE Transactions on Pattern Analysis and Machine Intelligence
A Survey on Content-Based Retrieval for Multimedia Databases
IEEE Transactions on Knowledge and Data Engineering
A Prototype Implementation of Metadata Generation for Image Retrieval
SAINT-W '04 Proceedings of the 2004 Symposium on Applications and the Internet-Workshops (SAINT 2004 Workshops)
Information Modelling and Knowledge Bases XV
Information Modelling and Knowledge Bases XV
Automatic Adaptive Metadata Generation for Image Retrieval
SAINT-W '05 Proceedings of the 2005 Symposium on Applications and the Internet Workshops
| Hi-index | 0.00 |
In this paper, we present the detailed theory and implementation on an automatic adaptive metadata generation system using content analysis of sample images with a variety of experimental results. Instead of costly human-created metadata, our method ranks sample images by distance computation on their structural similarity to query images, and automatically generates metadata as textual labels that represent geometric structural properties of the most similar sample images to the query images. First, our system screens out improper query images for metadata generation by using CBIR that computes structural similarity between sample images and query images. We have realized automatic selection of proper threshold-values in the screening module. Second, the system generates metadata by selecting sample indexes attached to the sample images that are structurally similar to query images. Third, the system detects improper metadata and re-generates proper metadata by identifying wrongly selected metadata. Our system has improved metadata generation by 23.5% on recall ratio and 37% on fallout ratio rather than just using the results of content analysis even with more practical experimental figures. Our system has its extensibility to the various types of specific object domains with inclusion of computer vision techniques.