Internet search engines are definitely well developed tools for finding textual information online. The same cannot be said for the equivalent image search engines that so far have been keyword based. A small Canadian startup called Incogna wants to change that having developed a new way to find images using a combination of keyword and image queries.
Incogna is the brainchild of Kris Woodbeck a University of Ottawa student who worked on image recognition during his graduate studies. Kris developed algorithms inspired by the function of the human brain trying to take advantage of the parallel processing capabilities of new digital computers includes hardware such as GPUs. This is how Kris describes the ICRE (Image Content Recognition Engine) that is the backbone of Incogna,
ICRE uses clusters of graphics processing units and Incogna's patent-pending distributed SIMD indexing algorithms to sort through your data. It's fast, efficient and is the visual data mining solution of choice.
Computer vision comprises some of the most complex, subtle and processor-intensive tasks in computing. The large scale of many visual databases further amplifies the difficulty of mining useful information from your data. Our technology uses methods derived from how cells in the visual cortex process information, faithfully represented on high speed GPUs, in order to extract and recognize the meaningful parts of any visual input.
Receiving the support of PRECARN and
NSERC, Incogna was born and after some alpha testing it opened to the world a couple of weeks ago.
Using Incogna, users begin searching for images using keywords. For example, one can simply search for images of the Eiffel tower. The search engine returns the best matches using keyword metadata information. Users can then refine their search by clicking on any of the returned images to find other images in the engine's index that are visually similar. Integrating this image-based search with images from a product catalog, Incogna also gives users a new way to shop online. You can try the image search engine
here.
I tried it for a while earlier today and I did get some very good results. I was actually rather impressed by it. The only problem is that so far only a small number of images have been indexed by Incogna's web robots. I wonder if the visual recognition engine will scale nicely as the number of images in the index grows to the many millions of images available online. Incogna says that quality of results is more important than the size of the index and I agree but if the index is small and users can't find what they are looking for, then they will go back to Google or other keyword-based search engines. I am really going to be keeping an eye on this company and their image search engine because I think they have a bright future ahead of them.