Increasing colour image segmentation accuracy by means of fuzzy post-processing

This paper presents a colour image segmentation method which attains a high segmentation accuracy even when regions of the image that have to be separated are very similar in colour. The proposed method classifies pixels into colour classes. Competitive learning with "conscience" is used to learn reference patterns for the different colour classes. A nearest neighbour classification rule followed by a block of fuzzy post-processing attains a high classification accuracy even for very similar colour classes. A correct classification rate of 97.8% has been achieved when classifying two very similar black colours, namely, the black printed with a black ink and the black printed with a mixture of cyan, magenta and yellow inks.