This paper describes an MIMD algorithm for the detection of constant curvature features in an image of man-made objects. This method involves intelligent partitioning of the edge points belonging to the object contour into logical divisions so that geometric token extraction algorithm for each partition can work independently. Partitioning is done using curvature information. A new technique is proposed for the estimation of curvature on the basis of local information. Modified Hough transform based techniques have been applied for the extraction of linear and curved segments from individual partitions. Experimental results have established effectiveness of the scheme.