A Scheme for Initial Exploratory Data Analysis of Multivariate Image Data

Klaus Baggesen Hilger, Allan Aasbjerg Nielsen, Rasmus Larsen

    Research output: Chapter in Book/Report/Conference proceedingArticle in proceedingsResearchpeer-review

    41 Downloads (Pure)

    Abstract

    A new scheme is proposed for handling initial exploratory analyses of multivariate image data. The method is invariant to linear transformations of the original data and is useful for data fusion of multisource measurements. The scheme includes dimensionality reduction followed by unsupervised clustering of the data. A transformation is proposed which maximizes autocorrelation by projection onto subspaces with signal-to-noise ratio dependent variance. We apply the traditional fuzzy c-means algorithm and introduce two additional memberships enhancing the textural awareness of the algorithm. Cluster validation is performed by examining the partition density of the segmentations as a function of the number of classes applied. Results are presented for a synthetic 2-band noise degenerated image and for an 8-band SeaWiFS scene.
    Original languageEnglish
    Title of host publicationProceedings of the 12th Scandinavian Conference on Image Analysis (SCIA)
    Publication date2001
    Pages717-724
    Publication statusPublished - 2001

    Fingerprint

    Dive into the research topics of 'A Scheme for Initial Exploratory Data Analysis of Multivariate Image Data'. Together they form a unique fingerprint.

    Cite this