Evaluación cuantitativa de la influencia de los espacios de color para la detección automática de células

Abstract

In this paper we study color spaces in the framework of automatic segmentation of cellular structures. In contrast to neuro-physical studies focused on human perception of color we resort to a quantitative evaluation of such impact using as reference segmentation outputs and ground truth images. The results show that color representation has a considerable influence in the capabilities of such algorithms and color spaces that separate chromaticity and luminance components, and normalize color differences, provide a representation of image data better than color spaces such as RGB and HSI used extensively in the field of biomedical image processing in the past.