Image Object Description Without Explicit Edge - Finding TR 91 - 049

The various tasks of computer vision dealing with objects, such as recognition, registration, and measurement, have typically required the intermediate step of finding an object edge, or equivalently the list of pixels in the object. This paper proposes a means for characterizing object structure and shape that avoids the need to find an explicit edge but rather operates directly from the image intensity distribution in the object and its background, using operators that do indeed respond to 11 edgeness 11 • The means involves a generalization of medial axis descriptions from objects defined by characteristic functions to those described by intensity distributions. The generalized axis is called the multiscale medial axis because it is defined as a branching curve in scale space. The result is stable to calculate and can be used to subdivide an image object into subobjects and detail subshapes as well as to characterize the shape properties of the objects, subobjects, and detail subshapes. The dominant train of thinking in object recognition, registration, and measurement has been that grouping is based on the local detection and tracking of edges. These edges or the regions enclosed by them'are first found. Then various measurements are made on the result, such as edge curvatures, medial axes, or moments of the object, and the final recognition, registration, and measurement are based on these. The difficulty of this approach is two-fold. First, from the point of view of physics, for an object in an image there exists no edge locus without a tolerance since the object can exist only via imaging and visual (here computer visual) measurements