Shape Description with a Space Variant Sensor: Algorithms for Scan-Path, Fusion and Convergence Over Multiple Scans
NEW YORK UNIV NY COURANT INST OF MATHEMATICAL SCIENCES
Pagination or Media Count:
One of the ways by which early human vision is sharply distinguished from machine vision is by the fact that the human visual representation is strongly space variant and that the human system builds up a representation of a scene through multiple fixations during scanning. In this paper, we discuss three algorithms related to the blending of a single scene from multiple frames acquired from a space variant sensor. 1 Given a series of space-variant contour based scenes, with different fixation points, we show how to fuse these into a single, multi-scan view, which incorporates the information present in the individual scans, 2 We demonstrate an attentional algorithm which recursively examines the current knowledge of the scene, in order to best choose the next fixation point, based on focusing attention in regions of maximum boundary curvature. 3 We discuss a simple metric for evaluating convergence over scan-path. This may be used to quantify the performance of 2 above, i.e. to compare the performance of various attentional algorithms. Finally, we discuss this work in the light of both machine and biological vision.