« Previous entry | Home | Next entry »

Tiltomo: a visual search engine

tiltomo.jpg
An interesting new site went live this morning. It's called Tiltomo, and it's a visual search tool that uses algorithms to find the similarity in images.

For now, the site is a proof-of-concept. The creators have pre-loaded the site with two databases of more than 280,000 images from Flickr (using Flickr's API). Users can search for images that are related by theme or subjects, or images with like color or texture. From the site's about page:

Our aim is to develop a Visual Search tool that makes it easier to find the images you like. Our approach is to allow free thinking, where search results are influenced by what you see. Our system encourages you to make a broad search and narrow the types of images you like. It's as simple as searching a keyword, then clicking the image you like. to see more. We feel this way of searching allows you to follow your own creative path, and find images that you might not normally look for.

We have been trading emails with the developers over the past couple of days. They explain the concept further thusly:

Our research has shown that most people look at images using three criteria. In order of importance, these are Color, Texture and in some situations Shape. Our Visual Search calculates separately both Color & Texture in varying amounts to determine what we call "Visually Similar." "Visually Similar" can be very subjective. For this reason we have two Visual Search modes. The first is "100% Color / Texture"; this matches images that are "similar in terms of only Color / Texture. Our second mode "Theme" uses Color / Texture and also takes Visual Context into consideration. This mode is most likely to match images that have a Visual / Contextual relationship.

The developers say that they see the technology being useful for "Consumer Image Sites, Stock Photo Libraries, Home Image Management Software. In fact anywhere a user needs to find a particular type of image that's hidden amongst a crowd." Indeed, we've already seen this type of concept used at sites such as iStockphoto, where you can search for images with similar colors.

There is other stuff in the works, including a Firefox plug-in that will allow "the user to visually search any image he/ she finds on the web against another image database."

So who's behind Tiltomo? We wish we knew. The developers have coyly refused to reveal themselves to us. If anyone knows, feel free to speak up.


Trackbacks
TrackBack URL for this entry:
http://www.siliconbeat.com/cgi-bin/mt331/mt-tb.cgi/1181

Links to blogs that reference this entry:

From: Visual
Visual
Excerpt: "Both" means that the property in question applies to both grid (for character grid devices or bitmap . vis...
Tracked: August 17, 2006 9:30 PM

Comments

This search engine is really interesting. I'm wondering how you can figure out patterns that well in graphics files.

So far, I think the results are pretty good.

Webmetricsguru on March 3, 2006 5:02 PM
Comment link

Undoubtedly Internet search will continue its development in diverse forms. Here is my vision of one of these forms.

A search that is oriented towards the physical nature of real entities. Selecting on the "physical nature of real entities" means being able to select the searched for entities based upon their components and the arrangement of those components into an organic whole. Implied by this is that that which is being sought has a physical presence ( e.g. a molecule vs a belief ).

Such a search would have a client interface that allows the searcher to easily construct a simple graphical model of what is being sought. This search returns links to Web pages, each of which presents an
identification, a graphic, and a discussion of the indexed entity. The results are returned in rank order. Rank is determined solely by the indexed entity's conformance to the physical model used as the search key.

In December of 2004 I became intrigued by this idea. I built my dream. It can be seen and used at http://siblab.com/SearchByDesign

Joseph Mansigian on March 3, 2006 8:10 PM
Comment link

There is also a new site for discovering and rating good images called imgSeek ( http://our.imgseek.net/ ).

You can tag images you see on the web and retrieve them later. And by rating images, the system learns and will then be able to recommend images you may like. It's also possible to browse images by shape and color similarity.

They are using the image similarity engine from an open source desktop app (http://www.imgseek.net/) also called imgSeek.

Here's an example of similar image search: http://our.imgseek.net/image/show/5250

PaulIrving on March 6, 2006 4:28 AM
Comment link

This is very intersting! But the search result is similar images rather then the content like Google Image search. It probably serves a niche.

Helen Wang on March 6, 2006 1:14 PM
Comment link
Post a comment












Remember personal info?