Segmentation of wood microanatomy images using multiscale classification

In this work we study wood in detail to understand better its nature, because of wood has a wide range of applications. There are wood laboratories that study microanatomic properties in samples using a microscope. We propose a method based on digital image processing which seeks to improve the analysis of these properties. In this initial proposal, we show a segmentation of wood microanatomy images for identifying microanatomy structures and to quantify their concentration in a sample. The approach consists on the creation of a scale space based on morphological operations and a clustering technique using the fuzzy c-means algorithm. The multiscale space is created by applying open-close and close-open morphological operations over images. The scale space is based on several shapes. Scales are determined through a granulometric study over the image by using Fourier transform. The segmentation is evaluated by considering images segmented by experts.