Computers learn to recognise objects
Researchers have developed a new computing technique which allows computers to use contextual information to identify the objects in an image.
To demonstrate the software, researchers from the University of California cite the example of a tennis game. Recognition software which is unaware of this context might identify the ball as a lemon, even if it correctly identified the racket and court.
However, a program with the contextual software would identify the racket and the court, deduce a tennis game and correctly reason that a lemon would not be in the picture and so identify it as a tennis ball instead.
The researchers say there are numerous possible applications for reliable recognition software, such as automatic tagging and organisation of images, and improved computer vision for robots used in manufacturing.
The research was presented at the eleventh IEEE International Conference on Computer Vision in Brazil.