Image segmentation algorithms applied to wood defect detection

• Published in: Computers and electronics in agriculture 2003-12, Vol.41 (1), p.157-179
• Main Authors: Funck, J.W, Zhong, Y, Butler, D.A, Brunner, C.C, Forrer, J.B
• Format: Article
• Language: English
• Subjects: algorithms; Biological and medical sciences; Defects; detection; Forestry; Fundamental and applied biological sciences. Psychology; Generalities. Production, biomass. Quality of wood and forest products. General forest ecology; image analysis; Image processing; Image segmentation; methodology; Quality of wood and forest products; Scanning; Wood; wood defects
• ISSN: 0168-1699; 1872-7107
• DOI: 10.1016/S0168-1699(03)00049-8

Image segmentation is a key stage in the detection of defects in images of wood surfaces. While there are many segmentation algorithms, they can be broadly divided into two categories based on whether they use discontinuities or similarities in the image data. Each algorithm can also be categorized based on other factors such as whether it uses color or gray-scale data and is a local or global operator. While this presents a wide variety of approaches for segmenting images of features on wood surfaces, it also makes it difficult to select the most appropriate techniques. This paper presents the results obtained from using a variety of algorithms for wood surface feature detection and defines several measures used for examining algorithm performance. A region-based, similarity algorithm that was a combination of clustering and region-growing techniques exhibited the best overall performance. This was particularly true for defects that are subtle, meaning they blend in with other natural features on wood surfaces that are not considered defects. Examples include blue stain, pitch streaks, and wane. The clustering with region growing algorithm improved the detection accuracy of pitch streaks by over 20 percentage points compared to the next best algorithm. However, if subtle defects are not of interest, the edge detection algorithms performed as well as the region growing algorithm but with slightly better clearwood detection accuracies. The influence of color information, local-basis analysis, and camera resolution on algorithm performance varied by segmentation technique and defect category. Because each wood processing application has its own unique set of defect detection requirements, conclusions regarding which algorithms and factors are best must be made in the context of those processing requirements.