Wednesday, September 1, 2010


Content-based image retrieval (CBIR) using ~1m flickr photos sourced from Cophir. Enter a query and Xyggy will find visually similar images based on their content. Improve relevance by dragging images in and out of the interactive Xyggy query box. Toggle images on and off from the search while looking. Go ahead and try it! Xyggy makes finding and searching an engaging and fun experience. Interact, explore and discover. Perfect for content search on media-rich sites and touch devices such as the iPod, iPad and Android. Content can include image, video and audio.

Access the Xyggy Images demo

* How does the UI work?

- enter keywords/phrases and results will be displayed.

- drag one or more photos (one at a time) into query box to find other similar pictures

- drag images in AND out of the query box.

- toggle photos on and off while in the query box. useful while finding photos of interest.

- keywords/phrase can also be toggled on and off.

* Does the content-based image search work? Yes, and here are some examples:

a. sunflower

- results for query "sunflower":

- drag photo to find other similar ones:,&sunflower

- drag another picture:,194630700,&sunflower

b. sunflower + pendant

- results for query "sunflower":

- drag non-sunflower photo to find other similar ones:,&sunflower

* But, why don't many of the queries return meaningful results?

Because, there aren't any visually similar photos in the flickr image data set to match the query.

* The performance is slow?

The service runs on a hosted virtual server in the US.

No comments: