Human Visual Perception and Spatial Models of Colour

Colour can deeply change its characteristics when put and analyzed into its spatial context. A growing family of algorithms treat/modify/enhance colour information in its visual context, also known as spatial colour methods. There is a long lasting debate about how to judge their performance. In our opinion this is a challenging task and still an open problem. Two main variables affect the final result of these algorithm: their parameters and the visual characteristics of the image they work on. We aim at discussing the visual configurations in which a spatial colour methods show interesting or critical behaviour. A survey of the more significant visual configurations will be presented and discussed. These configurations include phenomena, such as colour constancy, colour contrast, and colour assimilation. The discussion will present strength and weakness of different algorithms, hopefully allowing a deeper understanding of their behavior and stimulating discussions about searching a common judging ground. 1. THE NEED OF SPATIAL COMPUTATION Colour sensation is not directly linked to the spectral characteristics of the perceived signal. With the same triplet of stimuli, almost any colour sensation can be reproduced [1]. Context plays a very important role in changing colour appearance. Moreover, perceived signal has a poor channel interdifference, due to the high overlap of the red and green cone sensitivities (see Fig. 1). Fig. 2 shows a coloured scene acquired with two different spectral sensitivities, shown on the right side. The first is a typical film response (digital camera one is similar) with a very limited overlap, while the second is that of human cones (including pre-retinal absorptions), with a considerable overlap. From a signal detection point of view, the result of this big crosstalk is an image with poorly saturated colours. Using human spectral sensitivities the maximum red/green ratio possible is 2, not enough to fill the expected colour dynamic range of a standard 8 bit per channel image range. These pale colours contradict our everyday perception experience: the colours we perceive are not this pale. Spatial comparison probably originated to compensate the poor signal acquired by the retina. Fig. 1 – Human cone sensitivities. Thus, the signal starts from the retina with poor unsaturated colours and ends in our brain with the wide range and deep saturations that we are used to perceive. Is in the brain that the visual information is processed and enhanced. 0 : Spectrum (45) gm© 400 500 600 700 1.00 0.00 Sensitivity curves of the