A novel genetic algorithm for recognition of object shapes with movable parts

Computer vision techniques developed to date are effective in identifying images of rigid object shapes that are grabbed under different viewing positions. This is made possible by matching unknown object contours or their geometrical invariant descriptors against reference object models which had been previously stored in an image database. Despite the moderate success, little has been mentioned in how existing methods can be extended to recognise objects with moving parts. Apparently, the major problem is that the appearance of an object and its representation can change dramatically if some of its components are allowed to move. Hence given two dissimilar shapes, it is difficult to determine whether they originate from different objects, or images of the same item but with a certain part displaced from its original position. In this paper, a novel genetic algorithm for matching near planar object shapes with movable parts is reported. The method is invariant to Euclidean transformation, and is insensitive to noise contamination and distortions on the object boundary