Model-based partial shape recognition using contour curvature and affine transformation

The authors present a novel technique for recognizing partially obscured and overlapping objects by using the contour curvature and affine transformation. This technique is based on the use of an invariant attribute of an object, called a footprint, for the purpose of hashing. The recognition strategy is to match the given contour curve against all model objects and select the best matching. The recognition procedures are divided into reading each object picture, retrieving a footprint, building a model objects database, finding break points, and matching. Associated techniques, such as border tracking with chain code, setting a footprint hashing table, and connecting an orientational linked list, are also discussed