Viewnet Architectures for Invariant 3-d Object Learning and Recognition from Multiple 2-d Views

3 ABSTRACT 3 The recognition of 3-D objects from sequences of their 2-D views is modeled by a family of self-organizing neural architectures, called VIEWNET, that use View Information Encoded With NETworks. VIEWNET incorporates a preprocessor that generates a compressed but 2-D invariant representation of an image, a supervised incremental learning system (Fuzzy ARTMAP) that classiies the preprocessed representations into 2-D view categories whose outputs are combined into 3-D invariant object categories, and a working memory that makes a 3-D object prediction by accumulating evidence over time from 3-D object category nodes as multiple 2-D views are experienced. VIEWNET was benchmarked on an MIT Lincoln Laboratory database of 128x128 2-D views of aircraft, including small frontal views, with and without additive noise. A recognition rate of up to 90% is achieved with one 2-D view and of up to 98.5% correct with three 2-D views. The properties of 2-D view and 3-D object category nodes are compared with those of cells in monkey inferotemporal cortex.