2School of Information and Enviromental Sciences

Images of the human retina vary considerably in their appearance depending on the skin pigmentation (amount of melanin) of the subject. Some form of normalisation of colour in retinal images is required for automated analy-sis of images if good sensitivity and specificity at detect-ing lesions is to be achieved in populations involving di-verse races. Here we describe an approach to colour nor-malisation by shade-correction intra-image and histogram normalisation inter-image. The colour normalisation is as-sessed by its effect on the automated detection of microa-neurysms in retinal images. It is shown that the Naı̈ve Bayes classifier used in microaneurysm detection benefits from the use of features measured over colour normalised images. 1 introductio