Abstract
We apply new contour features: (1) Point features by computing the convexity and curvature in small contour neighborhoods. (2) Segment features by segmenting the contour into convex, concave and straight segments, and computing length and curvature measures for each segment. (3) Global features by computing the mean, maximum and minimum of all point and segment features. Features can be extracted from noisy contours with convex, concave and straight parts, but also from completely convex ones, for the purpose of shape analysis or identification (ID) tasks. Using only four global features, a nearest-mean classifier yielded a perfect ID rate of 100% on diatoms with minute differences in shape, which are difficult to identify even for diatomists.
Original language | English |
---|---|
Pages (from-to) | 1283-1285 |
Number of pages | 3 |
Journal | Pattern Recognition |
Volume | 37 |
Issue number | 6 |
DOIs | |
Publication status | Published - 1 Jun 2004 |
Externally published | Yes |