Automated editing and intelligent tagging

October 16, 2007 by pmelcher.

The future of photography is approaching fast. The fully automated photo agency is getting closer than we think and the cost of getting right image out will be dropping significantly.

Smart Tagging or kewording. A Penn State-developed software called TT or Tagging over Time analyzes the pixels of an image and returns a list of possible taggs. The user just needs to select the appropriate ones. Furthermore, the system learns and remembers interaction so that for future images with similar content, it becomes much easier to select the right tag. but that is not all. If the tagging of an image changes over time because of a change of perception, it will learn that to and change the taggs appropriatly. The authors of the software appropriately use the example of the twin towers.  In the past, they would have had the keywords “financial” or “business”. Now they are more likely to have “9/11″ or “terrorism”. All the image will retro actively be modified, based on user experience. It would probably have a hard time with conceptual words, although similar scenes or situation can trigger the same concept.

On the editing side, the same researchers have also created  a machine- learning component. To do this, the system uses visual features such as contrast, depth-of-field indicators, brightness and region composition from publicly rated photographs to learn the statistical models for high- and low-quality images. Now, just imagine if you could combine this with the technology of a Picscout or Idee. Rather than only going after copyright infringer, it would also learn from legitimate usage and return a huge collection of patterns and trends to be added to the learning algorithm. You would soon have an automated editing program. Great for microstocks, who spend most of their time, and money, manually filtering the images. But also incredibly useful for any photo agency receiving a lot of content. Furthermore, it could work on the photo buyer side, where a website could set up some pre define rules on a photo feed and publish a new image, every hour, automatically.

Funny how these technologies come out of Universities and research centers instead of the mega agencies. On would think that a Corbis, for example, instead of opening galleries on Second Life, would be working on cost saving, efficiency boosting applications. And the  Getty’s , instead  of creating  a  prosumer site for giggling teen agers , would jump on the opportunity to create a Getty lab instead.

The FAWM project I have started last year is all about finding and creating the right tools to minimize human interaction in the photography work flow. The digital age, up to now, has added more work, more steps into the “creation to licensing” process. It is time to put some of our intelligence and experience into the technological soup and come out with a new recipe for the perfectly streamlined photo agency.

 Article on smart tagging here

Posted in Search, keyword, corbis, getty, Microstock | Print | No Comments »