Description of contour with meaningful points

One of the most visually descriptive features for images is the contour of the object(s). In order to describe objects with lesser number of descriptors, linear or cubic curves are fitted to the contours of the objects. The end points of these finite length curves are usually meaningful spots on the contours. In the work presented here, edges are found by the Canny edge detector, followed by Principal Component Analysis (PCA) for determining corners and inflection points. These points are classified as dominant, corner, soft corner or inflection points. Curves are fitted to the sub-contours between successive such descriptive points on the object contour.