Image analysis of lignocellulosics

Abstract

For some decades texture has been an important object of study in the community of computer vision and image processing, yet more research is needed to make texture algorithms for pattern analysis and classification applicable in different areas. Texture is a key feature in many natural and synthetic images, and the process of texture analysis is an essential step in several image processing applications such as industrial inspection, remote sensing of earth resources, medical imaging, object recognition and content-based image retrieval. Numerous methods for texture analysis have been proposed in the literature. These methods involve extracting features and use different schemes to represent the selected features, which can be divided into five categories: structural, statistical, model-based, spectral and agent-based. Even with this variety of methods for texture analysis, all of them analyze texture globally, as if only one texture pattern in the image occurs. This project proposes a model of texture analysis based on the heterogeneity of patterns of texture in the same image. To perform this task, images of wood will be used, which is thanks to its hierarchical structure and heterogeneity suited for such analyses. The texture images to be analyzed are microscopic cross-sections of wood obtained by conventional light microscopy, as well as X-ray CT scanning, the latter in 2D and 3D. The proposed project can be divided into three work packages: wood recognition, orientation of fibers and tissue quantification in 3D. The first one concerns the classification of wood species by cross-sectional microscopic images of wood composites. The second deals with the analysis of wood, in this case specifically aimed at the orientation of fibers, influencing the wood's strength and durability. Finally, the third work package concerns tissue quantification in 3D comprising three-dimensional analysis of the different anatomical structures of wood. (AU)