From an industrial perspective proper characterization of thermo-mechanical pulp (TMP) fibres is most important. Scanning electron microscopy (SEM) and computerized image analysis have been one of the most applied techniques for assessing fibre structures. However, the quantification of fibre cross-sections requires intervention of a manual operator, which is subjective and time-consuming. Discriminant analysis is a statistical approach to obtain the discriminant functions which best separate two or more classes of objects or events. Although there are several discriminant analysis methods, we have proven that a Mahalanobis Discriminant Analysis (MLDA) is proper for classification of wood pulp fibre cross-sectional shapes. MLDA gives discriminant functions by hypersurface or curved lines taking into account data distributions. Such classification has been implemented in the Shape descriptor plugin, which performs an automatic classification of fibre cross-sectional shapes. SEM images containing a large amount of fibre cross-sections are classified automatically into e.g. intact fibres, splits, fibres with shoulders, discontinuity, and bundles of fibres. Such classification will ease the sub-sequent automatic quantification of fibre morphology, which will be demonstrated. In addition, the effect of effective resolution and image quality on the classification will be demonstrated.
Shape descriptor plugin, Discriminant analysis, thermo-mechanical pulp fibres, characterization, Mahalanobis distance
Presenting author: Asuka Yamakawa
Organisation: Department of chemical engineering, Norwegian University of Science and Technology (NTNU)
co-authors: Gary Chinga-Carrasco (Paper and fibre research institute (PFI))