Segmentación de imágenes de color imitando la percepción humana del color

Usually, related works on color image segmentation train neural networks with the colors of the image to segment, then the number of dominant colors within the image is obtained in order to process the image using fuzzy cmeans, where the colors are represented in the RGB space. The drawback with these methods is the neural networks must be trained every time a new image is given; but also, in the RGB space the color’s chromaticity can be altered by its intensity. In this paper we propose to segment the images using chromatic data of colors, by training a neural network with chromaticity samples of different colors, which can be employed to segment any image just training it just once; the number of colors the neural network recognizes depends on its size. We show experiments with images of the Berkeley segmentation database using competitive neural networks and self-organizing maps. 71 Research in Computing Science 114 (2016) pp. 71–81; rec. 2016-03-12; acc. 2016-05-15