SYSTEMS AND METHODS FOR TOPOLOGICAL REPRESENTATION OF GEOMETRIC CONTEXT IN AN IMAGE BY DISCRETE TEXTURE TRACES

摘要

Novel patch representations which are used in machine vision and are quasi-invariant to a large class of smooth deformations are provided. These patches are derived from a continuous domain model, called Profile Traces, which is a function only of the topological properties of the image patch and is, by construction, invariant to any homeomorphic transformations of the domain. The model is defined in terms of intersections of equivalence sets. A discrete-domain approximation called Discrete Texture Traces (DTT) is provided. DTT is designed to be computationally practical and is shown to be quasi-invariant to smooth spatial deformations. Methods and system that implement the steps of the methods are also provided.