Vision is the principal sensory system for mammals; 30% of the human brain is dedicated to vision. One of the most intriguing aspects of the mammalian visual system is the presence of massive feedback from higher centers to lower ones. In contrast, most of the current computer vision systems propagate information from lower levels of abstraction to higher levels. The architecture of future computational platforms can be improved by applying principles learned from these systems.
 
    We present a simple hypothesis of one principle, called Input Competition, which allows feedback from higher centers to influence processing at lower centers. In a model of the thalamo-cortical system of a cat, the cortical feature detectors compete not for the right to represent the output at a point, but for exclusive rights to abstract and represent part of the underlying thalamic input. We show how the principle of Input Competition is useful for detecting straight and curved lines that cross each other in images and to efficiently recognizing fingerprints.