20 most similar images are displayed as small thumbnails.
- If one of these was the desired image, the user simply clicks on that thumbnail to retrieve the full-sized image.
- Otherwise, if the desired image doesn’t appear among the thumbnails, the user may modify the query and “match” again.
- interactive retrieval: Every time the user pauses from painting for a moment, the application retrieves a new set of thumbnails.
- Quick speed: Median time to retrieve the target image appears to be around 20 seconds for a database of 1000 images.
- Criteria for quality:
- For painted queries, 51 of 87 test queries retrieved the correct image as a “best match”.
- For scanned queries, 66 of 89 test queries retrieved the correct image as a “best match”.
Other Application using this algorithm: imgSeek
imgSeek is a standalone image management application that incorporates that algorithm as well. The image you are looking for can be expressed either as a rough painted sketch or as another image you supply.
- Group your photos by similarity for easy browsing.
- Cluster automatically by color, date (group events automatically using an adaptive clustering algorithm for time differences), filename or image features.
- Find all duplicate images on your collection with the parameters you specify. (dimensions, filesize, filename, similarity, average luminance)