Object recognition: obtaining 2-D reconstructions from color edges
Gunter Bellaire, Kay Talmi, E. Oezguer, Andreas F. Koschan · 2002
This article presents a complete hybrid object recognition system for three-dimensional objects using characteristic views (ChVs). High-quality 2-D reconstructions have to be generated from intensity or color data, respectively, to integrate the ChV representation method into a recognition system. First, we present an approach for obtaining edges and gradient information from color images. Second, we present an adaptive edge linking process that transforms the image data into a suitable form. The edge linker uses an energy function to combine data from the edge detector and color gradient information to improve the results of the edge detector for the 2-D reconstruction. Results are presented for real color images.