Invariant object recognition via surface reconstruction neural networks
Jenq–Neng Hwang, Hongwei Li · 2003
Classifying objects that are distorted by similarity transform and detection/occlusion noise is a difficult pattern recognition task. The authors present a novel and robust neural network solution based on detected surface boundary points. The method operates in two stages. The object is first parametrically represented by a surface reconstruction neural network (SRNN) trained by the boundary points sampled from the exemplar object. When later presented with distorted object without point correspondence, this parametric representation reduces the effects caused by detection/occlusion and also allows the mismatch information back-propagated through the SRNN to iteratively determine the best similarity transform of the distorted object. The distance measure can then be computed in the reconstructed representation domains between the exemplar object and the aligned distorted object.>