The philosophy has been to maintain high spatial resolution within the CT images, in order to show the annual growth rings and a range of grade-reducing features. Previous tomographic assessment of logs of apparent uniformity show variable quality for sawn and veneer applications upon log breakdown, and this assessment correlates poorly with the end utilisation achieved.

CT images have been obtained for both hardwood, mainly eucalyptus regnans (mountain ash), eucalyptus delegatensis (alpine ask) and eucalyptus marginata (jarrah), and softwood (pinus radiata) samples using logs up to 450 mm in diameter. CT was used to reconstruct images of 512 x 512 12-bit pixel data sets using a 10 mm thick X-ray beam with a typical X-ray energy of 120 kV_peak and 170 mA beam current. The most common defects observed in the hardwood species and discussed in this paper are asymmetric or irregular growth patterns including wandering pith, tension wood, knots, cracks, rot and kino (gum) veins. CT images from pinus radiata specimens are characterised by high bulk density sap wood, a lower bulk density heart wood, a significant density variation between early wood and late wood, and knots.

A range of image processing techniques has been developed to automate the characterisation and location of these defects, and these are discussed in this paper. Images that have clearly defined growth rings allow pith location readily by finding zero- crossings following a second derivative operation. More compact growth ring structure requires the use of morphologic operators. Pith location is obtained with sub-ring precision by the histogramming of these processed images and the technique is reliable for logs of irregular geometries and sections containing knots and other defects.

Cracks are not always readily discernible by external inspection. In the CT images, air filled cracks are detected by thresholding methods, while cracks that are narrower than the CT image pixel dimension are tracked using morphologic close residues. IN a similar fashion rot is characterised by thresholding or morphology methods. Arc-volume methods are used for the detection of kino in hardwood samples, while knots can be identified by a number of methods derived from local density distribution within the CT image, or by subtraction methods of adjacent CT slices as the knots "move" in characteristic ways.

The relevance of this type of image processing using CT images of harvested logs to the Australian timber industry, and some of the remaining difficulties in implementing this kind of technology, will also be discussed in the paper.