A flat symmetric object viewed under perspective (orthographic) projectiorl yields a convergent (skew) symmetry. Unlike skew symmetry where the axes constrain the orientation of the underlying object to lie on a hyperbola in the gradient space, for convergent symmetry the object orientation can be uniquely determined from the axis and convergent point except for some special cases. However, detection of the axis and convergent point of a convergent symmetry has not yet received much attention. We present a technique which can detect the axis and convergent point of a convergent symmetric contmour in an image using distinguished points such as bitangents, corners, and inflection points on the contour. In addition to recovering the orientation of an object directly from the axis and convergent point of its convergent symmetry, when a model database is available, our technique is also able to recognize and locate the object by using the cross ratio as an indexing function. The orientation information is also used for verification purposes. Our technique is simple, efficient, and accurate. It is also robust to noise and partial occlusion. Experimental results on real images are shown.
Date of this Version