Free help and advice to the UK Further and Higher Education community

Helpdesk

Find me an image that looks like this one

Posted by Antony Theobald on Monday 26 October 2009 at 7:54am
Tags: content-based image retrieval | drawing | finding images | flickr |

Comment icon Comments (3)

UPDATE: The day after we posted this, GazoPa announced that it is now in open beta - available to the public without login. GazoPa is a new 'similar image search' service that finds images based on the visual content of any image you supply.

This means you can upload a photo from your computer, paste in the URL of any image online, or use a rudimentary drawing tool to create a basic sketch. GazoPa will then return a selection of similar images based on the colour, shape or layout of the image you supplied.

GazoPa is one of the newer sites experimenting with content-based image retrieval (CBIR) and is still in private beta, but if you are interested in trying it out you can request an invite by registering an email address - when I did this the request was accepted straight away. It is still in a beta testing stage but is now openly available to all.

Recent updates to the site include limiting search results to Creative Commons-licensed material from Flickr, with the option of filtering for pictures published within a certain time frame (e.g. 'today only', 'within a week', 'older than a year', etc).

Searching for images based on their visual content - instead of typing a term into a search box - is nothing new. There are several other examples of CBIR sites - some of which have been around for many years:

  • Webseek - Columbia University's system searches images and video from across the web
  • The State Hermitage Museum - uses IBM's QBIC (Query By Image Content) technology to search the collection
  • iStockphoto - the Advanced Search options include search by colour and layout
  • retrievr - lets you draw a sketch or upload an image and searches Flickr images only

The one thing they all have in common - and GazoPa is no exception - is the rather hit and miss nature of the search results. This can be frustrating, but can also lead to some interesting images that you may never have found using traditional text searching. Searching by colour often produces the most pleasing results, especially if your source image or drawing uses simple blocks of colour.

We talk about CBIR in a little more depth in our Review of Image Search Engines. Our review also points out that both Google and Exalead provide a form of CBIR with their face and colour recognition options.

Here's what GazoPa made of my fine art skills - the red image at the top is my sketch and the images below are based on its shape:

GazoPa screengrab

Comments (3)

1 of 3

Comment posted by Andy Coverdale on 26 October 2009 at 10:27am

Unlike GazoPa, which seems to work to a loose tolerance of image recognition, Tin Eye - a ‘reverse image search engine’ - seems to be remarkably effective at searching for instances of the exact same image on the web.

Check it out at http://tineye.com/

2 of 3

Comment posted by Antony Theobald on 26 October 2009 at 11:01am

Thanks for pointing that out Andy - TinEye is particularly useful for finding whether an image is being used elsewhere.

There’s also a TinEye browser plugin which lets you right-click any image on the web to find matching images, but note that like GazoPa it only searches a fraction of the images available online.

3 of 3

Comment posted by Antony Theobald on 28 October 2009 at 10:04am

Post updated - GazoPa is now in open beta and available to the general public without login.

Commenting is not available in this channel entry.

<< Previous entry: Access all areas
>> Next entry: Reduce RSS subscription clicks for Google Reader