Applications of neural networks to colour recognition

Abstract. Colour specification and matching are important tasks used in a number of different industries. However, current methods can employ very complicated, elaborate pieces of laboratory equipment which are not portable, such as spectrophotometers with combinations of calibrated light sources and a large number of narrow band filters. Simpler, portable devices are typically not sophisticated enough to give a high degree of quality control due to the less sophisticated light sources and the necessarily smaller number of filters employed. Lastly, techniques such as the `match by eye' method can lead to deceptive metameric matches. This paper outlines the beginning of the development of a colour measurement system which is portable, easy to use, and more accurate than other portable devices by virtue of an artificial neural network used to analyze the data. With specialized training of the neural network, it could even be used in applications such as teeth and skin colour measurements.