Does Automatic Tagging Have a Commercial Future?

ALIPR is a technology that automatically analyzes image content and suggests tags that can be appropriate to the image. Although it sounds impressive, it's also a very complicated technical task so the current version of ALIPR suggests many absolutely irrelevant tags. And ALIPR, like Google Image Labeler game which offers manual tagging, helps describing an image only as a single whjole without an ability to clearly refer to individual objects.

As I’ve already blogged on Riya and Polar Rose, companies that offer solutions for automatic facial recognition, it would be unfair to not mention another product targeted towards the same market. It is called ALPR which stands for Automatic Linguistic Indexing of Pictures. It was officially launched on November 1, 2006 and is currently available at http://www.alipr.com/.

When I was blogging on Polar Rose and Riya, I mentioned that photos are not only about people so these technologies can be hardly useful if I need to find, say, an inanimated object. ALIPR seems to be an answer to that challenge.

Do you remember the Google Image Labeler game? In that game, you randomly paired with someone who is using the same service. Then you see an image and have to provide as many labels as possible to describe each image you see. When your label matches your partner's label, you'll earn some points and move on to the next image until time runs out. The game’s goal was to involve Web surfers to tagging images so Google would be able to improve search results.

ALIPR tags images alone and automatically. This is how it works: you upload an image to the ALIPR sit, click a button, ALIPR analyzes the image content, and displays a list of tags that could be appropriate for the image. ALIPR offers API for software developers and search engine companies to integrate automatic indexing capabilities to their applications.

Then I tried to upload an image with the developers’ notice in mind that the current version of ALIPR “is not designed for black&white photos, manipulated photos, objectionable images, cartoons, sketches, framed photos, etc.” This is the photo I’ve uploaded:

This is the list of tags I’ve got:
Ocean, sport, dogsled, water, man-made, wind, people, ship, surf, boat, winter, sail

Well, I realized that this is only first version of the software and it should learn much, but actually I expected at least one word closely related to a bridge. Another picture I tried was:

And the tags look as follows:
Indoor, drawing, ancestor, man-made, animal, people, art, photo, thing, decoration, ship, tool, decoy, steam, bird

It's not too bad for a machine recognition, but there is much work to do. Frankly speaking, I doubt it has good chances to become a commercial application in the nearest future although it definitely has a great scientific potential.

Another problem is that ALIPR, like Google Image Labeler game works with the image as a whole without an ability to clearly refer to individual objects on the image. After all, if a photo contains multiple objects and you or smart program describe each of them by adding a tag, how one who found this photo can correlate the tags with the particular objects on the photo? Although tags that describe an image as a whole are important, why not to go further and describe the content within the photo:

The object-specific tags can be stored with the image wherever it is located or moved (it goes without saying that an ability to see the original image, without any tags is required). They can feed both a Web surfer and search engine with much more information about the found image content. For you, as an end-use, such information means a new opportunity to get some knowledge. For a search service, it means an ability to index every object within the photo individually. It could be even highlighted on the found photo in the same way as Google highlights text fragments in the found text documents.

So the word “automatic” in conjunction with “image indexing” can work in some magic way for potential investors, it does not necessarily mean success. What we all need are:

1. An easy mechanism/tool for tagging content within the images and adding general tags to an image. The tags should be added manually, by an image owner (Web site owner).
2. An approved standard recognized by the search services for storing these tags in a way that lets them keep together with the image regardless of its location. Tags should travel together with the image wherever it goes.

Posted on February 28, 2007 by Alex Masycheff

See all posts