Artificial vision systems can not process all the information that they receive from the world in real time because it is highly expensive and inefficient in terms of computational cost. However, inspired by biological perception systems, it is possible to develop an artificial attention model able to select only the relevant part of the scene, as human vision does. This paper presents an attention model which draws attention over perceptual units of visual information, called proto-objects, and which uses a linear combination of multiple low-level features (such as colour, symmetry or shape) in order to calculate the saliency of each of them. But not only bottom-up processing is addressed, the proposed model also deals with the top-down component of attention. It is shown how a high-level task can modulate the global saliency computation, modifying the weights involved in the basic features linear combination.
