As with any other technology, image recognition has its limitations. Pushing the boundary of what’s possible and what’s not is our primary task at iTraff Technology. Take a look at our case study.
In most situations it works flawlessly, but one can always think of conditions so extreme that it will fail to do so Now, after several years of development we have a system that is able to recognize objects in situations so difficult that most people consider it impossible. This post describes one of them.
Many users are concerned about the requirements for the size and quality of images. It is true that to get the best results one should provide high-resolution reference images as well as accurate query pictures, but let’s not demonize. If you have high-quality images – use them, you will have mostly reliable results. But keep in mind also that our API is powerful enough to recognize images properly even in very harsh conditions with minimal relevant information. Here’s an example.
Have you ever considered if it would be possible to recognize an image as small as a thumb nail… and partially occluded? Well guess what, recognize.im does so effortlessly.
Recently I got hold of DK’s 2012 catalog and decided to test for myself just how well our technology performs in recognizing the products presented within it. On page 44 I found this tiny thumbnail for “First Aid Manual+”:
To check if I wasn’t just lucky I turned some pages and found this even smaller thumbnail of “Star Wars Reader”:
This time I covered even more of its surface with my clumsy finger. This is what I got:
Actually, this time the algorithm wasn’t sure which one of the Star Wars series it was, so it matched three different volumes. This isn’t particularly surprising since almost all of the cover’s content was obscured save for the logo, which is common to the series.
The icing on the cake is that the above experiment was performed against a reference database of over 250k book covers.
As you can see one doesn’t have to print A0 posters or even full-page advertisements to make them interactive. Big and small alike, iTraff Technology can recognize them with a high degree of confidence. Size really doesn’t matter – as long as you have a clear photo we will tell you what’s in it.
Believe it or not, similar rules apply to the other parameters of our image recognition, for instance size and the compression level of the reference and query images. Big images are great, but you will be surprised how much we can squeeze out of crappy ones. More examples will follow – stay tuned.