© Copyright 2008 by T. Pavlidis

Most of the sites listed below are BETA versions, therefore they may eventually achieve better results. What is shown below corresponds to their state in November to early December 2008. If any owner of a site believes that my description is inaccurate, they should send me e-mail (see my index page) and I will post their answer here.

Citations not otherwise identified are to the main paper "Limitations of CBIR".


GazoPa is a system that is being currently developed by Hitachi. One can apply for an account at http://www.gazopa.com/sign_in and upload pictures of his/her choice to test the BETA version of the system. I tried several images and I obtained reasonable results for only the two shown below.


The matches returned by GazoPa are shown below. The 9th returned image is actually quite a close match to the query, the only time I have encountered a close match in all the tests I have performed with all accessible sites. Several of the other returns match the overall shape of the query. The second return is unrelated but one can see how the choice was made.

Screendump of GazoPa screen for palm.jpg

In the case of the mosque, the two first returns as well as the 6th to 9th also seem to match the query, although not as close as in the case of the palm tree.

Screendump of GazoPa screen for mosque.jpg

It seems that if an image contains a well defined shape GazoPa does a decent job, certainly better than any other system I have tested. Whether that performances can be repeated for other major characteristics remains to be seen.

There is aldo the issue of whether such broad matchings are good enough for practical purpose. If a user wanted to find images of mosques (as it is likely to be the case) rather than simply towering structures the answer is not satisfactory.

Gazopa returned less than satisfactory unswers in all the other tests I tried (more than 20). Two of them have been included in the oral presentation.


The image below shows the results of the site http://www.tiltomo.com/. The site does not allow user submitted images only rankings relative to one of its selected images. The second image seems to be close to the first bit the rest of the matchings seem to be irrelevant.


TinEye by Idée, Inc.: I submitted several images and I always received the reply "0 matches". The following messages are also displayed with the returns:

  • "TinEye looks for the specific image you uploaded, not the content of the image. TinEye cannot identify people or objects in an image." And
  • "Our search index is still very small—just a fraction of all the images on the web! But our index is always growing, so be sure to search for this image again later."

Apparently, TinEye only looks for duplicate images on the web.

imgSeek: There is no online demo. A user must download software on his/her machine. The imgSeek site claims the following:

  • "imgSeek is a photo collection manager and viewer with content-based search and many other features. The query is expressed either as a rough sketch painted by the user or as another image you supply (or an image in your collection). The searching algorithm makes use of multiresolution wavelet decomposition of the query and database images."

Given the poor experience I had with other wavelet based systems, I did not pursue the issue further. (See Appendix A about my experience with another wavelet system: Retrievr.)

LTU: The site does not allow for submissions. They also claim to be using Color Palettes, not an encouraging approach (see the end of Section 3) or slide No. 17 in the ICPR08 presentation.)

Imagewiki: "Internal Server Error -- The server encountered an internal error or misconfiguration and was unable to complete your request. ..."

IMEDIA: "Error on Page"

VIDEOSURF: It searches only by tags, no CBIR (in spite of rumors to the contrary).

SnapTell: It works only with i-phones and I do not have such a gadget..