Before image analysis was done, image noise had to be reduced in all image
stacks. Then the image stacks were virtually cut into subvolumes of certain size.
Before starting the analysis the images were segmented to gain binary information
(0 and 1). For MDF the Otsu method [9] showed high potential in dividing
the image in material and void. For OSB and PB an image segmentation algorithm
based on the analysis of variance was chosen, which was already presented
elsewhere [10]. This procedure enabled to distinguish between void, low
density regions (e.g. early wood) and high density regions (e.g. latewood).
In order to obtain the pore size distribution of the investigated panels procedures
of mathematical morphology were used [11]. The main procedures are erosion
of a set of voxels I (1) with a certain shaped structuring element H, followed by
dilation (2) with same structuring element. Both procedures together are called
opening (3). With this procedure it is possible to find regions which fit to the
structuring element.