Metrical evaluation of spatial content for segmentation­based image retrieval

Abstract

Interest in the image content analysis has been motivated by expansion of imaging in manifold scopes of activity, the availability of large image libraries, growth of multimedia applications, etc. One of the ways for partial elimination of a semantic gap between low-level visual features and high-level human concept is to analyze spatial properties of image parts specifically induced by segmentation. To provide a region-based image retrieval with a query ‘ad exemplum’ (to wide extent), a new metric to compare arbitrary nested partitions is proposed. Studied metrical properties of partially ordered quotient sets provide e.g. the objects search independent on the background.