Multi-resolution


In biological vision systems, particularly referring to primates, broadly speaking the light sensitive elements on the retina are distributed in a non homogeneous way so that a central area is densely covered while the periphery has a smaller number of sensitive elements. Many authors argued that this acuity distribution on the retina provides a hardwired implementation of scale invariant analysis for object recognition. Moreover, in this way the subject perceives a global view and then only the interesting/relevant components of such view are observed by redirecting the eye so as to expose the fovea (central area having the maximum resolution) to analyze those components.
This mechanism may also switch between different resolution levels (excluding the extremes); such switching also reflects a hypothesis-testing modality which is typical of a biological reasoning process that maximizes the use of limited physical resources. At each analyzed level, the total amount of processed information is constant since the number of picture elements which are handled is approximately the same regardless of the image resolution.
Multi-resolution representations have been proposed for object recognition since they provide the framework for emulating the focus of attention strategy typical of biological systems. In the Laboratory the problem of matching two-dimensional objects described by multi-resolution representations have been considered. Each object is modeled as a tree, in which nodes correspond to boundary segments and arcs connect nodes at successive levels of resolution. The children of a given node describe the structural change occured to a given segment between consecutive resolution levels. A two-dimensional object can be recognized as an instance of a model possibly rotated and translated, if the corresponding tree can be mapped into the tree.


Back to IST Home Page