Abstract
Sapstain is considered a defect that must be removed from processed wood. So far, research in automatic wood inspection systems has been mostly limited to dealing with knots. In this paper, we extract a number of colour and texture features from wood pictures. These features are then assessed using machine learning techniques via feature selection, visualization, and finally classification. Apart from average colour and colour opponents, texture features are also found to be useful in classifying sapstain. This implies a significant modification to the domain understanding that sapstain is mainly a discolourization effect. Preliminary results are presented, with satisfactory classification performance using only a few selected features. It is promising that a real world wood inspection system with the functionality of sapstain detection can be developed.