[Back]

I. Janusch, W. Kropatsch:
"Shape Classification according to LBP Persistence of Critical Points";
Poster: 21st Computer Vision Winter Workshop (CVWW2016), Rimske Toplice, Slovenia (invited); 2016-02-03 - 2016-02-05; in: "Proceedings of the 21st Computer Vision Winter Workshop", Slovenian Pattern Recognition Society, Ljubljana, February 2016 (2016), ISBN: 978-961-90901-7-6; 12.

This paper introduces a shape descriptor based on a com-
bination 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 cap-
tures 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.

Shape descriptor · Shape classification · Local topology · Persistence · LBP · Local features

http://publik.tuwien.ac.at/files/publik_252828.pdf