Abstract
In a previous work, we empirically analysed the so-called "relevant colours" in paintings. These relevant colours describe the number of colours that stand out for an observer when glancing at a painting. On average, 19 relevant colours were identified, enabling reliable segmentation of colour images via a small palette of colours. The purpose of this study was to analyse how observers with different colour deficiencies (specifically, 11 deuteranopes, 11 protanopes, 3 deuteranomalous and 4 protanomalous observers) perceive the most representative colours in paintings. To achieve this, a set of painting images was presented to these observers. They were instructed to mark those areas within the image that contained a relevant colour via a mouse. The observers visually scanned the entire painting and selected pixels belonging to relevant chromatic areas. The study included 20 images from the Prado Museum in Madrid, Spain. The results revealed that, as expected, the average number of relevant colours identified by all observers was relatively small, consisting of only 9 colours. This number was lower than the average number of colours found for observers with normal colour vision in the earlier study and reveals that dichromatic observers experience a degradation in their mental impressions of the observed paintings. Nevertheless, the results indicate that there is almost no difference in the number of relevant colours obtained between the two types of colour vision deficiencies; deuteranopes seem to be able to use more relevant colours than protanopes need for the same painting (10 and 8 relevant colours, respectively). By analysing the segmentation of pictorial scenes with relevant colours identified for each painting as colour seeds, we inferred how colour-deficient observers perceive and communicate much of the critical chromatic content in a painting. Through maximizing communicative effectiveness and minimizing the cognitive load, our results assess the colour information retained by anomalous observers, indicating that they lose approximately 20% of the representational content compared with normal observers.