Minimum description length with local geometry

Martin Styner, Ipek Oguz, Tobias Heimann, Guido Gerig

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

    Abstract

    Establishing optimal correspondence across object populations is essential to statistical shape analysis. Minimizing the description length (MDL) is a popular method for finding correspondence. In this work, we extend the MDL method by incorporating various local curvature metrics. Using local curvature can improve performance by ensuring that corresponding points exhibit similar local geometric characteristics that can't always be captured by mere point locations. We illustrate results on a variety of anatomical structures. The MDL method with a combination of point locations and curvature outperforms all the other methods we analyzed, including traditional MDL and spherical harmonics (SPHARM) correspondence, when the analyzed object population exhibits complex structure. When the objects are of simple nature, however, there's no added benefit to using the local curvature. In our experiments, we did not observe a significant difference in the correspondence quality when different curvature metrics (e.g. principal curvatures, mean curvature, Gaussian curvature) were used.

    Original languageEnglish (US)
    Title of host publication2008 5th IEEE International Symposium on Biomedical Imaging
    Subtitle of host publicationFrom Nano to Macro, Proceedings, ISBI
    Pages1283-1286
    Number of pages4
    DOIs
    StatePublished - 2008
    Event2008 5th IEEE International Symposium on Biomedical Imaging: From Nano to Macro, ISBI - Paris, France
    Duration: May 14 2008May 17 2008

    Publication series

    Name2008 5th IEEE International Symposium on Biomedical Imaging: From Nano to Macro, Proceedings, ISBI

    Other

    Other2008 5th IEEE International Symposium on Biomedical Imaging: From Nano to Macro, ISBI
    CountryFrance
    CityParis
    Period5/14/085/17/08

    Keywords

    • Correspondence
    • Image registration
    • Image shape analysis
    • Modeling
    • Statistics

    ASJC Scopus subject areas

    • Biomedical Engineering

    Fingerprint Dive into the research topics of 'Minimum description length with local geometry'. Together they form a unique fingerprint.

  • Cite this

    Styner, M., Oguz, I., Heimann, T., & Gerig, G. (2008). Minimum description length with local geometry. In 2008 5th IEEE International Symposium on Biomedical Imaging: From Nano to Macro, Proceedings, ISBI (pp. 1283-1286). [4541238] (2008 5th IEEE International Symposium on Biomedical Imaging: From Nano to Macro, Proceedings, ISBI). https://doi.org/10.1109/ISBI.2008.4541238