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

The problem of recognizing partially occluded parts is of considerable interest in the field of manufacturing automation. In this paper, we describe a technique of recognizing partially obscured and overlapping two-dimensional objects by using the contour curvature and affine transformation in a model base. This technique is based on the invariant attribute of an object, called footprint, for the purpose of hashing. By such a recognition method, we could identify an object by matching all model objects against a period of the partial contour of composite objects and by voting one of the best matches from the pre-established model base. The technique is divided into the following procedures: digitization from camera, thresholding into binary, object labeling, border operator, chain coding, footprint extraction, building the model base and hashing table, finding breakpoints, building the segment table, affine transformation and matching process.