Abstract
An evaluation strategy for visual key image retrieval systems is proposed in order to show the design criteria of a querying interface on mobile devices. Indexes (lists of visual keys) generated by different number of visual keys and visual features are validated using Art- Explosion 600,000, which contains about 300 semantic categories and over 100,000 natural photos. The result suggests that access to a collection with a visual key can provide a relevant image in rank 10 and 4 relevant images in rank 20 when the number of visual keys is 60, which is the lower limit. In portable devices, which display 16 visual keys per page, users can at least access a required image by browsing only 4 pages with 60 visual keys, and can use the image for related subsequent queries by using the other image retrieval functions.
Original language | English |
---|---|
Pages (from-to) | 713-722 |
Number of pages | 10 |
Journal | Journal of Advanced Computational Intelligence and Intelligent Informatics |
Volume | 16 |
Issue number | 6 |
DOIs | |
Publication status | Published - Sept 2012 |
Externally published | Yes |
Keywords
- Clustering
- Image retrieval
- Indexing
- Retrieval accuracy
- Visual feature