Contextual visual search
|Product:||The Empora Fashion Search Engine|
|Functionality:||Image search engine; online advertising platform|
The company has applied patent-pending scientific research into image search and retrieval in a fundamentally different way to other exponents of web search technology, allowing users to browse and search for relevant items using images rather than keywords. In essence, PIXSTA enables an image to be submitted as a search term, generating results in the form of a set of visually similar images. Lately PIXSTA got transformed into Empora Ltd which is enhancing the technology and adding a social layer on top of the visual search technology.
The PIXSTA image search engine is the result of several years of academic research by Dr Daniel Heesch, a graduate of St John’s College Oxford and Imperial College London, UK.
The search engine does not rely on tags that may, or may not have been, added to images; instead, it analyses images down to the pixel level, to identify visual attributes like shape, texture, colour and the kind of objects they contain. PIXSTA’s technology compares images directly at this visual level rather than at the textual level, making tags unnecessary.
Like other search companies such as Google and Yahoo, the challenge facing PIXSTA is the profiting from its technology. Its first attempt at making a profit is in the form of an online advertising network for fashion retail. In this network is predominantly PIXSTA's own websites - however a number of fashion media brands seem to have piloted or are piloting PIXSTA's solution in the shopping sections of their websites. The database of images comprises the online collections of multiple fashion retailers – effectively, advertisers. The system is presently a ‘closed’ system, in that search term images must be an image from the database; submitting any such image brings up other images from the database as results.
Conceptually similar to Google AdWords, PIXSTA AdImages generates revenue from the advertisers in the form of click-throughs, which take the visitor directly to the appropriate product page on the retailer site. When on a media brand's website, Click-through revenue is shared between PIXSTA and the media brand. Equivalent to Google AdSense, Pixsta's 'ImageSense' allows publishers to price contextual pages via image matching technology. Ads displayed are heavily targeted by the understanding of visual attributes. Due to the high contextual attributes profitability rates are rumoured to be higher than Google AdSense.
PIXSTA evidently revealed early deployments in November 2006, but didn’t make any further formal statements until May 2008. At this time it announced several live pilot deployments, and claimed more to come.
The company alluded to further online advertising networks in other industries where goods are sold on their visual appearance, including shoes, art, antiques and home furnishings.
The company also claimed that the PIXSTA image search engine is sufficiently accurate and sophisticated to enable its use in facial recognition applications.
In April '09 PIXSTA launched Empora.com a fashion discovery website with style and colour search, in which they realized pixta's unique technologies.
No timescale was given for any product evolution.
- Image Search for the World Wide Web (2009) Invited Lecture (IET London, 20th Jan)
- Empora.com UK clothing & accessories web site
- Empora.de German clothing & accessories web site