In this paper, we deal with the colour naming visual task in computer vision. Our goal is to develop a colour naming model to be included in a real surveillance application, where images have to be automatically annotated with people clothes description, for a further content-based image browsing.Although colour naming has been a usual goal in psychophysical research, it is a quite novel topic in computer vision. Colour naming is posed as a fuzzy set problem where each colour category is presented as a fuzzy set with a characteristic function. Our goal is to find a model which provides membership values as similar as possible to the values that would give a real observer.To this end, we propose a Sigmoid-Gaussian model as the membership function of the colour fuzzy categories. We analyse its properties and results to confirm the suitability of this model versus most common Gaussian models. To test the results, we have developed a colour naming experiment that has provided a set of membership values for a set of colour samples. Although the data used is far from being a complete learning set, it has been a first step to evaluate the proposed model.