Comparison of segmentation methods for digital image analysis of confocal microscope images to measure tracheid cell dimensions

Abstract

Image analysis is a common tool for measuring tracheid cell dimensions. When analyzing a digital image of a transverse cross section of wood, one of the initial procedures is that of segmentation. This involves classifying a picture element (pixel) as either cell wall or lumen. The accuracy of tracheid measurements is dependent on how well the result of the segmentation procedure corresponds to the true distributions of cell wall or lumen pixels. In this paper a comparison of segmentation methods is given. The effect of segmentation method on measurements is investigated and the performance of each method is discussed. We demonstrate that automated segmentation methods remove observer bias and are thus capable of more reproducible results. The contrast for confocal microscope images is of such quality that one of the fastest and simplest automatic segmentation methods may be used.