Publications (2)0 Total impact
ABSTRACT: Features for a new application in pattern recognition are presented. The objects to be classified are coccoliths (marine microfossils). Since the objects differ both in their outline and in their internal structure the features developed take into consideration both kinds of variability.
The features concerning the internal structure are computed on the greyvalue images produced by a scanning electron microscope. These features include measures of both of the angular variation, using the autocorrelation function, and of the radial variation within the objects by an approximation with Legendre polynomials.
The features concerning solely the shape of the objects include the content of long and short straight segments of the outline, and measures based upon ellipse fits and the smallest bounding rectangles to the outlines. In addition, moments are calculated on the objects' binary filled interior.
04/2006: pages 853-858;
ABSTRACT: A hierarchical approach for the classification of marine microfossils is presented. A-priori knowledge about the sought shapes is used for the image segmentation. The objects are divided into elliptical and hammer-like objects. For the search for elliptical objects RANSAC is combined with a genetic algorithm resulting in a robust and relatively fast algorithm for ellipse detection. The found ellipse directly yields important features for the subsequent classification. On the other hand, for the detection of hammer-like objects a hierarchical structural approach is employed. The proposed method can handle both occlusion and missing object parts.