Measuring rectilinearity

Abstract

Two new methods for computing the rectilinearity of polygons are presented. They provide shape measures and estimates of canonical orientations which can be used in applications such as shape retrieval, object classification, image segmentation, etc. Examples are presented demonstrating their use in skew correction of scanned documents, deprojection of aerial photographs of buildings, and scale selection for curve simplification. Furthermore, testing has been carried out on synthetic data and with human subjects to verify that the measures do indeed produce perceptually meaningful results.