ABSTRACT
Until recently, statistical analysis of similarity shape from images was re-
stricted to a small amount of data since the appearance of similarity shape
is relative to the camera position with respect to the scene pictured. In this
chapter, we study the shape of a 3D configuration from pictures of this config-
uration in 2d images without requiring any restriction for the camera position-
ing respect to the scene pictured. This methodology uses standard reconstruc-
tion methods from computer vision. In absence of occlusions, a set of point
correspondences in two views can be used to retrieve the 3D configuration
of points. A key result due to Faugeras (1992) [111] and Hartley, Gupta, and
Chang (1992) [143] states that two such reconstructions differ by a projective
transformation in 3D. Sughatadasa (2006) [322] noticed that the object that is
recovered without ambiguity is actually the projective shape of the configura-
tion. This cast a new light on the role of projective shape in the identification
of a spatial configuration.
