COLOUR IMAGE SEGMENTATION IN PERCEPTIVE COLOUR SPACE

Abstract

A new colour image segmentation algorithm in perceptive colour space has been proposed in this thesis. We have analysed the perceptive colour model - Hue, Saturation and Intensity (HSI), and its conversion methods from Red-Green-Blue (RGB) colour space. The YUV-HSI colour space is found to be closest to human's perception, and is selected in the development of our colour image segmentation algorithm. The powerful colour recognition capability of human beings in discerning colour differences and in utilising fuzzy information for object recognition drives us to develop a fuzzy rule-based colour image segmentation algorithm - Fuzzy Hue Mapping algorithm. Beginning with the definition of a number of colour-fuzzy set, all the colour pixels in an image are mapped to a colour number with fuzzy membership function values. Using information from the neighbouring pixels, a "Voting" algorithm is exercised to classified pixels into definite colour number defined in the HSI colour space. The algorithm has been tested with standard colour images, traffic scene images and pictures of colour traffic sign with different degrees of success.