The authors draw upon their previous contributions to the fields of 3D object recognition and 3D object reconstruction to obtain new methods of increasing the robustness of systems that use these technologies.

Two types of new measures are introduced. The first, view likelihood, measures the probability that a certain view of a specified 3D object can be observed. The second, view stability, measures the degree of change when a perturbed viewpoint is considered. The authors show that both measures are identical if the prior distribution of camera orientations of a specified object is uniform, all viewpoints being equally likely.

The material is presented clearly in six sections and three appendices. After an interesting introduction, the second section reviews and compares previous work in the field. The stability and the likelihood of views are discussed in sections 3 and 4, where the properties of these measures are investigated. The fourth section describes various applications of view likelihood in object recognition and 3D reconstruction. The three appendices describe affine and similarity distances, calculating view likelihood and stability, and maximum likelihood reconstruction.

This research material will be of value to readers interested in 3D object recognition and reconstruction. All of the references are useful, and the length of the paper is just right.