Evaluation of Graph Partitioning Measures for Perceptual Organization


Grouping low-level image features based on innate organization among them, without specific object model knowledge is known as perceptual organization. Based on these low-level features, a graph is created which has in it Gestalt inspired relations. This graph is partitioned to form salient feature clusters. The general K-way partitioning of the graph can be approximated by recursive bi-partitioning, where at each step the graph is split into two parts based on a partitioning measure. In this study, three such measures are compared. These measures are studied extensively both theoretically as well as empirically, details of which can be found in the relevant publications.