Bayesian Model-Based Automatic Landmark Detection for Planar Curves

Justin Strait, Sebastian Kurtek; Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2016, pp. 86-94

Abstract


Identifying landmarks, points of interest on a shape, is crucial for many statistical shape analysis applications. Landmark-based methods dominate early literature; more recently, a method combining continuous shape outlines with landmark constraints was proposed. Unfortunately, methods requiring landmark specification depend on the number selected and their locations; such annotations are tedious for large datasets and subject to human interpretation. This work provides a Bayesian model-based method for automatic landmark selection, based on good approximations of landmark set interpolations. We outline an appropriate prior and likelihood, allowing for efficient posterior inference on landmark locations. The model allows for location uncertainty quantification, an important inferential procedure for further analysis. A method for selecting an appropriate number of landmarks is also discussed. Applications include a simulated example, shapes from the MPEG-7 dataset, and mice vertebrae.

Related Material


[pdf]
[bibtex]
@InProceedings{Strait_2016_CVPR_Workshops,
author = {Strait, Justin and Kurtek, Sebastian},
title = {Bayesian Model-Based Automatic Landmark Detection for Planar Curves},
booktitle = {Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
month = {June},
year = {2016}
}