Reasoning with entropy graphs for image operators

摘要

A framework is proposed for structuring of large sets of image operators. The set of image operators is organized into a metric space. The metric measures the information content of one image operator relative to another by the entropy function on the partitions that are induced on a set of reference images. A non-symmetric version of this metric is proposed as an edge-sensitive quality measure for quantitative evaluation of compression and restoration algorithms. The metric provides a graph structure that can be utilized for classification of image operators. The image elements can be viewed as operators acting on the set of image operators by evaluation. The procedures developed for classification of image operators can hence be transferred to image segmentation procedures.