I have been investigating a couple of image recognition services that will return a description from an uploaded image. About a month ago I put one of these services, IQ Engine’s Smart Camera API, to use when building a demo app as part of a pitch for an office supplies company. The app allowed the user to quickly snap photos of objects and it would automatically create an inventory of supplies organized by category. The app took a photo, re-sampled it, and uploaded it to the web-service which would, in a short period of time, return a description of the photo. It became clear to me in testing this app that the image recognition would attempt to dynamically categorize the image. If a photo contained a product logo or other distinguishing mark a result was returned almost instantly. But, most of the time the images were not recognized instantly are were presumably put in a pool of images that appeared to be tagged by humans. (I noticed many spelling mistakes and inconsistent results with the same images). Regardless of how it worked, the response time was usually a few minutes and the tags were good. Employing a simple but effective system to best match descriptions to a set of predefined categories based on scoring matches in an array (The same technique used by the heroic Subservient Chicken), we were able to make an app that actually worked pretty well at categorizing photos of office related items. Here is a video of the app in action.