Visual search companies began licensing their technology to stock photo companies several years ago.  Though its promoters claim visual search use continues to grow in popularity throughout the industry, however, they also lament visual searches still account for a small percentage of all the searches performed on sites that offer it.  Keyword searching still makes up the vast majority of image searches performed on-line.

How it works:  For the most part, visual searches rely on a reference image, which the user must provide, usually by performing a keyword search first.  The user executes a visual search based on the reference image chosen from the keyword search results.  Visual search software typically converts digital images into numerical values which measure the various characteristics of the pixels, not only separately but also in relation to one another.  Paul Bloore, CTO at Idee, Inc. of Toronto, the visual search provider best known to the stock photo industry, lists the attributes as "color, texture, shape, background, luminosity, objects, and over 200 others."  The software tabulates these attributes from the reference image and then seeks out and retrieves images with similar dominant attributes.  Using Idee's software, a user can simultaneously enter a keyword, thereby narrowing the results to only those images that meet visual criteria and also bear the desired keyword in their metadata.  Some stock photo companies who use visual search software have this visual-with-keyword-search combination embedded in their systems.
 
The implementation of visual search technology in a given database of images requires the indexing in advance of the entire database in accordance with the system of numerical values.  Bloore says Idee's software indexes images at a rate of 20 per secjavascript:void(0); Collapse Extended textond, meaning it takes about seven hours to index a database of 500,000 images.

The developers of visual search software create proprietary algorithms for arriving at the numerical value of images in order to retrieve images that appear visually similar when viewed by the human eye.  These proprietary algorithms distinguish one visual search software from another.  To try out several different versions of this type of visual search, click on the following links:

http://www.masterfile.com (visual search technology from Idee, Inc.)
http://www.mediabakery.com (visual search technology from VIMA)
http://corbis.ltutech.com (visual search technology from LTU Technologies)
http://www.matton.com (visual search technology developed in-house by Matton Images)
 
San Francisco based CogniSign LLC, a relative newcomer to the stock photo visual search scene, has developed a different approach to visual searching.   The company recently released Xcavator, a software that allows the user to choose the desired characteristics of the reference image by clicking on them with a mouse.  The Xcavator system accepts up to six clicks on the reference image which the software will then seek to match.   For example, if one had a reference image of mountains and sky with random clouds, and one wanted to see the same image, but with more clouds, one could click on the mountains and the clouds, but not the blue sky.  

Micro-stock site, iStockphoto, uses two visual search features which they developed in-house.  One, called Color Profile, searches for images that simply match the color profile of the reference image.  The second, which, according to iStock’s Marketing VP Kelly Thompson, ranked highest among favorite features in a recent client survey, is called Copy Space.  That feature allows users to choose a section or sections of an image where they would like blank space for text placement.

Overall, however, on-line image researchers rarely use the visual search features available to them.  This perplexes those who have gone to no small expense to offer visual search, hoping for a competitive advantage.  Those companies report that once a client tries visual searching, they usually continue to use it.  One theory put forth for the relatively low usage of visual search features on stock photo web-sites is that several conditions have combined to prevent image researchers from adopting the new technology.   Over the years since image research went on-line clients have become accustomed to (and gotten very good at) searching by verbal description.  At the same time, keywording techniques have become more sophisticated. Finally, clients are reluctant to attempt new techniques for accomplishing a task when the technique they normally use gets the job done efficiently.  Clients gladly stick to keywords because that’s what works for them.

Leila Boujnane, the President of Idee, stresses the importance of visual searching to be used in combination with keyword searching, not as a substitute for it.  “We put forward the position that keyword searching is here to stay and that visual search enhances it but does not in any way replace it. We don’t view visual search as a replacement to anything that is being used today but merely as an enhancement – something used to improve the results from simple keyword searching.”


Please Complete Our Visual Search Survey

To get a better understanding of how clients view visual search features, we at abouttheimage have decided to conduct a survey about it.  We encourage any stock photo company who currently offers visual search technology on their own site or is considering doing so, to ask their clients to complete the survey.  Our survey does not call for respondents to provide any identification other than their occupations and we will share the results in this space when the survey is complete. Click here to go to the abouttheimage survey on visual search.