Recognition of Partially Occluded Elliptical Objects using Symmetry on Contour
June-Suh Cho, Joonsoo Choi · 2007
In this paper, we have discussed how to estimate parameters and to reconstruct the occluded shape of partial elliptical objects in image databases. In order to reconstruct occluded shapes, we used mirror symmetry, which provides powerful method for the partial object recognition. Unlike the existing methods, a proposed method tried to reconstruct occluded shapes and regions within objects, since most objects in a domain have symmetrical figures. However, we have limitations in the shape of objects and the occluded region of objects. For example, if a pan has an occlusion in handle, it cannot correctly reconstruct and be recognized. Another minor limitation of a proposed method is that it is sensitive to the pose of an object. For example, if we cannot see an ellipse due to the object's pose, we cannot recognize the object. After estimation, we have applied inputs, which include estimated parameters, to the existing classification trees, to get to the best matching class. All experiments are performed based on the classifier in earlier work. In experiments, the results show that the recognition of the occluded object is properly reconstructed, estimated, and classified, even though we have limited to the size of samples. In addition, we have experienced the power of the symmetry through experiments.