Researchers at the Fraunhofer Heinrich Hertz Institute (HHI; Berlin, Germany) have developed a system that can automatically analyze the content of images and assign keywords to them.
For each category of images to be tagged, a training set of images is first presented to the system, which contains relevant and irrelevant samples of the types of images to be categorized.
From these, key features such as edges, color, and geometric shapes are automatically extracted from the images.
In the second step, the system is trained to identify which key features extracted from the training set represent an image that would place it in a particular category.
The system is then able to determine if a previously "unseen" image belongs to a specific category or not by analyzing its features.
-- Posted by Vision Systems Design
