The Colour Group (Great Britain)Meetings 2005-2006

Prof Janos Schanda
Do LEDs need a new colorimetry?
With the use of LEDs in architectural and general purpose lighting, discrepancies between colorimetric and visual matches have been realized. Also the CIE colour rendering index has been questioned, as differences between calculated colour rendering indices and observed colour preference have become apparent.
The paper will deal with the problem of updating the CIE colorimetric system and the colour matching functions to provide measurement results that correspond better to average human colour perception. This problem was less apparent as long as broad-band light sources were used, but has now become critical with narrow-band LED sources.
Special emphasis will be given to the question of colour rendering, where one has to distinguish between the criteria of colour fidelity, colour preference and colour discrimination. For colour fidelity a measurement algorithm can be suggested that relies on current colorimetric knowledge. Colour preference and colour discrimination are more complicated, and the paper will show examples of how preference varies with the observed scene, the ability of observers to discriminate between small colour differences, and how observer expectations may influence their decisions. Directions for further research will be outlined.

Qasim Zaidi
Cortical decoding of shapes and colours: costs and benefits of expanding neural dimensionality
It is well established that retinal images are parsed in visual cortex by neurons tuned to many different orientations and spatial-frequencies at each retinotopic location. Similarly, the three colour signals transmitted by the retina and LGN are processed by cortical neurons whose preferred stimuli provide a much finer sampling of hue directions, and whose tunings progressively narrow along the visual stream. Taken together, these neurons provide high-dimensional representations of orientation, spatial frequency and colour. I will discuss the costs and benefits of expanding neural dimensions of visual processing. For many objects, information about 3-D shape in retinal images is carried by orientation flows and frequency gradients formed by projections of surface texture and isophotes of reflected light. Parallel processing of multiple orientations and scales, coupled with cross-orientation inhibition, permits flows and gradients to be extracted despite the presence of other orientations at each retinal location. In addition, objects and scenes are defined by elongated distributions in color space. Detecting objects in their natural habitats and separating illumination changes from material variations, requires distinguishing between color distributions at different orientations. Performance in these tasks is facilitated by the multiplicity of cortical colour cells, because mutually orthogonal colour distributions can be processed by mutually exclusive sets of neurons. I will conclude by showing how high-dimensional neural spaces enable natural materials to be embedded as simple surfaces, allowing linear discriminants to classify materials accurately and rapidly.