Contour Description Through Set Operations on Dynamic Reference Shapes

Abstract

Eight novel features for irregular shape classification which use simple set operations to compare contours to regularized reference shapes are introduced. The features’ intuitive simplicity and computational efficiency make them attractive choices for real time shape analysis problems such as defect inspection. Performance is evaluated through a brute force feature combination search, in which KNN classification rates of the proposed features are compared to several existing features also based on contour comparison. Results indicate that combinations of the proposed features consistently improve classification rates when used to supplement the feature set. Efficacy of the individual features ranges greatly, but results are promising, especially for Outer Elliptical Variation; its strong performance, in particular, calls for further investigation.