Title :
Contour profiling by dynamic ellipse fitting
Author :
Ciobanu, Adrin ; Shahbazkia, Hamid ; Du Buf, Hans
Author_Institution :
Vision Lab., Univ. of Algarve, Faro, Portugal
Abstract :
We present a new method for characterising 2D shapes, specifically elliptical ones, that can be successfully applied to any star-shaped object. Rather than using the classical centroidal profile approach, we dynamically fit an ellipse to the contour points and compute the distances of every point on the contour to the two ellipse foci. The sum of the two distances minus the length of the ellipse´s major axis, as a function of the angle for one full rotation, is the new contour-selection. This profile is normalised with respect to the object orientation by taking the orientation of the fitted ellipse´s major axis as a reference. Our method can perform better if a contour has defects or is partially occluded, it has no problem with very elongated objects, and it can also cope with objects with identical shapes but different sizes
Keywords :
biology computing; edge detection; feature extraction; object recognition; pattern classification; 2D shape contour; centroidal profile; contour profiling; diatom; dynamic ellipse fitting; feature extraction; object recognition; shape classification; Charge coupled devices; Digital cameras; Digital images; Equations; Laboratories; Object recognition; Optical microscopy; Pattern recognition; Shape; Valves;
Conference_Titel :
Pattern Recognition, 2000. Proceedings. 15th International Conference on
Conference_Location :
Barcelona
Print_ISBN :
0-7695-0750-6
DOI :
10.1109/ICPR.2000.903653