AbstractHere human trichromats were presented with two types of scenes – geometric and real-world scenes –tinted with a shade of colour in order to destabilize the perceived illumination, and chromaticity of the retinal image of each scene. Each trichromat was instructed to adjust the chromaticity of the object embedded within each scene until its surface appeared devoid of any hue in the DKL colour space which spans two chromatic opponent axes – the S–(L+M) and L– M axis – and a luminance axis – the L+M axis. The following observations were made : (i) across scenes, adjustments were dispersed along the S–(L+M) axis, along which daylight is known to vary; (ii) across trichromats, for the geometric scenes, adjustments were biased towards the S pole of the S–(L+M) axis for one group (group 1), and towards the (L+M) pole for the other group (group 2); (iii) for the real-world scenes, adjustments for both groups systematically converged towards the (L+M) pole. These results suggest that when the core set of priors upon which the human visual system typically relies become ill-equipped, the human visual system is able to recruit one of the two illumination priors – PriorS or PriorL+M – in combination with the representation it has formed over time about the spectral composition of the illuminant associated with scenes the trichromatic observer is currently being exposed to within its ecological niche, as it attempts to stabilize the chromaticity of the retinal image of real-world scenes.
A Multi-Scale Image Enhancement Model using Human Visual System Characteristics