On the usefulness of attention for object recognition

Today’s object recognition systems have become very good at learning and recognizing isolated objects or objects in images with little clutter. However, unsupervised learning and recognition in highly cluttered scenes or in scenes with multiple objects are still problematic. Faced with the same issue, the brain employs selective visual attention to select relevant parts of the image and to serialize the perception of individual objects. In this paper we demonstrate the use of a computational model of bottom-up visual attention for object recognition in machine vision. By comparing the performance of David Lowe’s recognition algorithm with and without attention, we quantify the usefulness of attention for learning and recognizing multiple objects from complex scenes, and for learning and recognizing objects in scenes with large amounts of clutter.