Shape classification according to LBP persistence of critical points

Abstract

This paper introduces a shape descriptor based on a combination of topological image analysis and texture information. Critical points of a shape’s skeleton are determined first. The shape is described according to persistence of the local topology at these critical points over a range of scales. The local topology over scale-space is derived using the local binary pattern texture operator with varying radii. To visualise the descriptor, a new type of persistence graph is defined which captures the evolution, respectively persistence, of the local topology. The presented shape descriptor may be used in shape classification or the grouping of shapes into equivalence classes. Classification experiments were conducted for a binary image dataset and the promising results are presented. Because of the use of persistence, the influence of noise or irregular shape boundaries (e.g. due to segmentation artefacts) on the result of such a classification or grouping is bounded.