SEGMENTACIÓN DE ELEMENTOS ANATÓMICOS EN IMÁGENES MICROSCÓPICAS DE MADERA USANDO TÉCNICAS DE VISIÓN ARTIFICIAL.

The characteristics of the elements of wood microscopic images are keys to identify the species of which comes a sample. This paper presented a novel segmentation method for pores, rays and fiber. The proposed methodology uses a set of algorithms from digital image processing consistent with the characteristics of each element. For pores, a texture descriptor, a clustering algorithm and active contours were used; for radios, mathematical morphology and Gabor filters were used for emphasis and to segment proceeds in three steps: mean-shift clustering, a Gibbs sampler and area opening; the fiber was taken as the surplus of previous segmentation processes. Experimental results with 18 manually segmented samples of Gaiadendron punctatum species demonstrate the efficiency of the proposed method, using performance statistics such as the overall success rate and the specificity and sensibility by element. [ABSTRACT FROM AUTHOR]