Recognition of 3D nonrigid objects by learning view change transformations

In this paper, we propose a method to recognize 3D nonrigid objects based on learning from examples without special knowledge on the target objects. We show the mappings from an arbitrary view to the standard view; and its rotated view can be synthesized even for a nonrigid object by interpolating examples given in the learning phase. Using such view change transformations obtained by learning, we can synthesize different views from a given input view. We show that the nonrigid object can be recognized by examining the consistency between the input view and the synthesized views.