Image pre-processing algorithms for isolation of defects in Douglas-fir veneer

Graduation date: 1987The goal of this research is to develop automatic techniques for\ud identifying defects in Douglas-fir veneer. To achieve this goal,\ud computer vision techniques, in particular image-processing, will be\ud applied. Several pre-processing steps may be required to enhance\ud images in making further analysis easier, faster, and more reliable.\ud Pre-processing is often necessary when analyzing natural\ud materials due to the amount of complex variation present. However, if\ud such analysis is associated with several compute-intensive passes over\ud a large data set, the possibilities of what can be achieved, is very\ud limited.\ud It is advantageous to reduce the data in a way that little or no\ud information is lost that may affect the image analysis. This reduced\ud image is then used in subsequent analysis to determine the actual type\ud of defect, its extent and exact location. Whether such a reduction\ud is possible usually depends on the intended application. For Douglas-fir veneer sheets at the plugging stage, the material is relatively\ud clear, with defects occupying only a minor portion of the surface. It\ud will be sham that reduction by pre-processing is indeed feasible.\ud This study investigates three potential image sweep and mark\ud (ISM) pre-processirg algorithms that perform this kind of reduction.\ud These algorithms are very diverse in their methods of operation and\ud include methodologies based on first-order statistics, mathematical\ud morphology, and color cluster analysis. It is shown in the\ud conclusions of this thesis that one of these algorithms is capable of\ud reducing the data of a typical Douglas-fir veneer image by 80% with\ud an error rate of only 0.3%. It is also shown that color is a\ud significant factor to be considered for the detection of defects