Performance analysis of an improved tensor based correspondence algorithm for automatic 3D modeling

3D modeling of a free-form object involves the acquisition of multiple views (range images) of the object to cover its entire surface. These views are then registered in a common coordinate basis by establishing correspondence between them. Our tensor based correspondence algorithm can automatically establish correspondence between overlapping view pairs and register them. In this paper we present an improved version of our tensor based automatic correspondence and registration algorithm and integrate it with a global registration and a surface reconstruction algorithm to make an efficient, automatic and complete 3D modeling framework. We also present a quantitative analysis of this automatic correspondence algorithm based on our results according to the following criteria: robustness to resolution, efficiency, robustness to the amount of overlap and noise.