Abstract
We present an information-theory approach to the selection of salient characteristics in an image. We conjecture that the choice of the specific features perceived as salient can be explained on the basis of the amount of usable information they provide for coding the entire image, and the amount of computing resources required in the visual system. If this conjecture is true, the vision primitives may be the product of a natural selection process amongst all conceivable visual patterns, based on information content and computational cost. These ideas were tested in a practical implementation with samples of natural images.