Computers just keep getting better and better, but for most tasks they still need human input. Human input needs time, and oh well, we don’t have a lot of time to spare in our days.
Have you ever surfed the net looking for an image that contains things that you want ? You might of noticed that you can search pictures by tags, but those tags can be fake and/or incomplete. And from the other side of the gun, when people post pictures they don’t usually have the patience to cover all the tags. And you spend at least half an hour if you want to find an image that contains all the elements you want.
But a team of researchers from Penn State developed a nice little thing called “Automatic Linguistic Indexing of Pictures in Real-Time (ALIPR)”. This is a statistical approach to photos that can make our task of finding photographs on the Internet much easier.
This system works like this : you take a bunch of pictures, you put in all the tags (patience my friend) and then the system scans those images for similar patterns amongst keywords. ALIPR studies the colors, the shapes, the sizes amongst those similar tags. When this development stage is finished, the system will be able to identify tags in a new picture. It’s more like training the computer than programming it to do something, much like the voice recognition programs. This is a big step from the way current image search engines work : analyzing the words around the image and the name of the image itself.
The team currently holds the patent for a previous version of this approach, called ALIP, and now is in the process of patenting the new and improved ALIPR.
Jia Li, a professor of statistics at Penn State further explains : “Our basic approach is to take a large number of photos — we started with 60,000 photos — and to manually tag them with a variety of keywords that describe their contents. For example, we might select 100 photos of national parks and tag them with the following keywords: national park, landscape, and tree,”"We then would build a statistical model to teach the computer to recognize patterns in color and texture among these 100 photos and to assign our keywords to new photos that seem to contain national parks, landscapes, and/or trees. Eventually, we hope to reverse the process so that a person can use the keywords to search the Web for relevant images.”
While the team is brainstorming and implementing new ideas into this program, the final selection will still need a human eye, but the variety will be much more precise : “There are so many images out there and so many variations on the images’ contents that I don’t think it will be possible for ALIPR to be 100-percent accurate,”
Leave a reply