Object registration for visual inspection operations

Describes the positioning module of a visual inspection system for dimensional verification. The system operates in two phases: learning and test. In the learning phase, the operator performs a set of inspection operations on the image of the prototype piece, while in the test phase the system processes the image of a piece that belongs to the class considered in the learning phase. For this purpose, the position of the test piece has to be determined. The authors represent the prototype and test pieces by attributed relational graphs (ARG) in which the nodes describe closed contours while the arcs represent their mutual relationships. Positioning is obtained by means of a graph matching technique based on a hypothesize and verify cycle. The authors use a hierarchical approach for graph matching; first they look for node correspondences (operating on a contour matching level), then they find the graph match by hypothesizing an arc correspondence and evaluating its consequences on the whole graph