3-D shape recovery by incorporating context-a connectionist approach

In this paper, we present a connectionist approach to the recovery of a 3-D object composed of known primitives from a 2-D image by incorporating context. A certain-to-uncertain scheme is proposed for extracting aspect coverings from a face graph of the image recognized. On the basis of the scheme, we employ Hopfield nets to verify candidate aspect instantiations (AINs) and rank order the likelihood of all AINs. Finally, a Hopfield net recovers the 3-D object based on the aspect covering.