Recently there is a growing interest in the use of colour for an increasing number of industrial machine vision tasks, particularly in the areas of quality control. The reason being that colour images can convey more information than greyscale representations. This is intuitive: while a colour can have only a single intensity value, an intensity value can correspond to any of a number of possible colours. However the main problem arises because there is always three times as much data to process in a colour image, while there is not necessarily three times as much information content.
In order to complicate matters further, there are numerous different ways of representing coloured images. Consequently each of the different schemes reflect different properties about the way colour is perceived. For example the Red, Green, Blue (RGB) Intensity representation of colour images attempts to model colours in terms of the theory of Trichromacy; the idea that a colour can be represented by a mixture of the three primary colours of Red, Green, and Blue. Other examples of colour representations include the YIQ, and YUV schemes for transmission of television signals. Although these methods are based on the notions of information theory for the optimal use of transmission channels, however the spaces produced by these techniques result in an approximation of the coding used in the optic nerve.
The focus of this particular area of research is therefore a full consideration of the problems of colour images. Reducing the quantity of data involved to levels approximating the quantity of information, and the examination of the different representation schemes for their suitability to particular image processing tasks, particularly image segmentation.
Image segmentation is a key area in the field of machine vision, since it forms an essential part of the process of relating objects in the real world with their representations in digital images. Colour being one of the more obvious characteristics of objects is considered to be of particular importance in the image segmentation process, and algorithms are being developed to use colour effectively in the segmentation process.
However many of the problems of image segmentation are basically philosophical: what constitutes an object? This point may seem like an epistemological quibble, but it does have a significant bearing on the segmentation problem: if there is no clearly defined notion of an object, then there can be no clearly defined notion of segmentation. This means that the goals of the task are difficult to specify, and its success difficult to measure. An important part of this study is the development and evaluation of suitable criteria for the establishing the success of image segmentation algorithms.
The research involves close colaboration with the Vision and Robotics Group at the University of Murcia, Spain.