Automatic visual inspection of wood surfaces

Abstract

A prototype software system for visual inspection of wood defects has been developed. The system uses a hierarchical vector connected components (HVCC) segmentation which can be described as a multistage region-growing type of segmentation. The HVCC version used in experiments uses RGB color vector differences and Euclidean metrics. The HVCC segmentation seems to be very suitable for wood surface image segmentation. Geometrical, color and structural features are used in classification. Possible defects are classified using combined tree-kNN classifier and pure kNN-classifier. The system has been tested using plywood boards. Preliminary classification accuracy is 85-90% depending on the type of defect.<>