This work gathers theoretical and experimental evidence showing the feasibility of a computer vision system aimed at performing view-based recognition under space-variant conditions. An artificial retina preserving the essence of space-variant sensing while passing around some of its intrinsic difficulties is described and baptised the square-zone retina. A new bottom-up attentional scheme, the generic attentional scheme, is tailored to the properties of the square-zone retina. The attentional strategy is designed to assist visual learning tasks under uncontrolled illumination conditions. Recognition theories are the focus of a review describing the progressive emergence of the view-based concept. The basic idea of that concept is adapted to show experimentally the interpolation properties of new object representation. A review on trans-saccadic memory and integration, as well as insights gained experimentally, have guided the selection of the nature of the image features extracted to construct the object representation. Finally, (i) the square-zone retina, (ii) the generic attentional scheme, (iii) the adaptation of the view-based concept, and (iv) the object representation are bound together to demonstrate the feasibility of view-based recognition under space-variant conditions.
展开▼