Inference of Stochastic Graph Models for 2-D and 3-D Shapes

摘要: Shape interpretation methods that model a shape using stochastic graphs can recognize many classes of nonrigid objects, even if the objects are partially occluded, and interpret complete scenes composed overlapping shapes. These also identify most parts each shape. This paper describes use graph as for class 2-D or 3-D shapes, presents learning infer models their symbolic primitives from examples. methods, well graph-covering method used scene interpretation, criterion minimum description complexity which eliminates need subjective parameters. One practical application this work is trainable real-time system recognizes hand gestures in images.