Robust active appearance models with iteratively rescaled Kernels

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

Active appearance models (AAMs) are widely used to fit statistical models of shape and appearance to images, and have applications in segmentation, tracking, and classification of structures. A limitation of AAMs is that they are not robust to a large set of gross outliers. Using a robust kernel can help, but there are potential problems in determining the correct kernel scaling parameters. We describe a method of learning two sets of scaling parameters during AAM training: a coarse and a fine scale set. Our algorithm initially applies the coarse scale and then uses a form of deterministic annealing to reduce to the fine outlier rejection scaling as the AAM converges. The algorithm was assessed on two large datasets consisting of a set of faces, and a medical dataset of images of the spine. A significant improvement in accuracy and robustness was observed in cases which were difficult for a standard AAM.

Bibliographical metadata

Original languageEnglish
Title of host publicationBMVC 2007 - Proceedings of the British Machine Vision Conference 2007|BMVC - Proc. Br. Mach. Vis. Conf.
PublisherBMVA Press
DOIs
Publication statusPublished - 2007
Event2007 18th British Machine Vision Conference, BMVC 2007 - Warwick
Event duration: 1 Jul 2007 → …

Conference

Conference2007 18th British Machine Vision Conference, BMVC 2007
CityWarwick
Period1/07/07 → …