Proposes an orthodox method of extracting curvature features based on a curve fitting approximation. It enables us to obtain analog values of curvatures of a given curve. This means that a so-called gray zone between two categories can be identified and also that very shallow concavities can be detected. For this purpose, cubic B-splines are obtained using a least squares method with natural conditions at the end-points. The method was tested on synthesized noisy data sets such as 2/spl rarr/Z, 4/spl rarr/9 and 1/spl rarr/3. The results are so good that the method can be used to obtain analog features as intended. Demonstrative experimental results are shown for the data set for 1/spl rarr/3.
展开▼
