Abstract

Despite the great demand for blood smear analysis in Brazil and worldwide, relatively few efforts have been directed to the automation of this important problem. This paper presents a prototype of a semi-automatic leukemia diagnosis program. emphasizing basic stops in pattern recognition as segmentation, filtering, and feature extraction. A supervised learning process segments blood smear images into four regions of interest: nucleus, cytoplasm, erythrocytes and plasma according to their color. Then, the measurements are performed, both general such as perimeter, area, factor for, circularity as well as innovative measures as curvature, skeletons, and multiscale fractal dimensión, which can provide more objective subsidy for diagnosis.