Unsupervised Shape Classification of Convexly Touching Coated Parts with Different Geometries

Visual inspection system of various manufactured geometries is widely adopted in industry whereas the inspection system's intelligence is yet to be improved. Mixing of various geometries is a common problem in coating industry and the inspection is carried out manually. This paper gives a framework for classifying the mixed parts from the images captured without prior information of geometries. The parts were segmented from the image using Otsu method followed by morphological operations. Then the borders were extracted and smoothened by Fourier approximation. The touching objects were separated using curvature analysis. Features such as area and skeleton were extracted from the individual parts. The geometries were then classified by k-means clustering successfully. The developed algorithm works for a variety of geometries and is independent of translation and rotation of the parts. © 2013 © 2013 CAD Solutions, LLC.