Rytis Stanikunas

Investigation of color constancy with a neural network

The perceptual stability of an objects color under different illuminants is called color constancy. We created a neural network to investigate this phenomenon. The net consisted of one input channel for the background and one for the test object. Each channel had a set of three (L, M, and S) receptors that were transmitting to three opponent neurons. The signals from the opponent neurons were passed to hidden neurons, which were connected to the output neurons. The output signal was generated from the three components of a color vector. The neural net was trained to identify the color of Munsell samples under various illuminants using the back-propagation algorithm. Our study investigated the properties of a successfully trained neural network. Based on the cross-neuron weight analysis, we report that the successfully trained neural net calculates color differences between the test object and the background. By comparing the human visual system to the neural net, we conclude that to satisfy the color constancy phenomenon, the human visual system has to contain two separate components: one to approximate the background color and the other to estimate the color difference between the object and the background.

Firelight LED Source: Toward a Balanced Approach to the Performance of Solid-State Lighting for Outdoor Environments

We report on a blue-amber (“firelight”) cluster of light-emitting diodes (LEDs) with extra-low correlated color temperature (~1860 K) optimized for outdoor lighting under mesopic conditions. When compared with common white LEDs, the firelight LED cluster shows considerably reduced indexes of melatonin suppression and skyglow, increased retinal illuminance for elderly people, but a reduced performance of perceiving colors, which, however, can be tolerated at mesopic luminance. In comparison with an almost metameric high-pressure sodium lamp, the cluster exhibits a potentially higher luminous efficacy, similar reaction time and detection threshold of luminance contrasts for achromatic targets, and noticeably improved color discrimination characteristics.

Colour matching of isoluminant samples and backgrounds: A model

A cone-opponent-based vector model is used to derive the activity in the red-green, yellow-blue, and achromatic channels during a sequential asymmetric colour-matching experiment. Forty Munsell samples, simulated under illuminant C, were matched with their appearance under eight test illuminants. The test samples and backgrounds were photometrically isoluminant with each other. According to the model, the orthogonality of the channels is revealed when test illuminants lie along either red-green or yellow-blue cardinal axes. The red-green and yellow-blue outputs of the channels are described in terms of the hue of the sample. The fact that the three-channel model explains the data in a colour-matching experiment indicates that an early form of colour processing is mediated at a site where the three channels converge, probably the input layer of V1.

Systematic violations of von Kries rule reveal its limitations for explaining color and lightness constancy.

Cone contrast remains constant, when the same object/background is seen under different illuminations-the von Kries rule [Shevell, Vis. Res. 18, 1649 (1978)]. Here we explore this idea using asymmetric color matching. We find that von Kries adaptation holds, regardless of whether chromatic constancy index is low or high. When illumination changes the stimulus luminance (reflectance), lightness constancy is weak and matching is dictated by object/background luminance contrast. When this contrast is masked or disrupted, lightness constancy mechanisms are more prominent. Thus von Kries adaptation is incompatible with lightness constancy, suggesting that cortical mechanisms must underlie color constancy, as expected from neurophysiological studies [Zeki, Nature 284, 412 (1980); Wild, Nature 313, 133 (1985)].

Colour and brightness shifts for isoluminant samples and backgrounds

Changes in colour appearance were studied under eight different illuminants. The most informative conditions were when sample and background were both contourless and isoluminant. Matching contourless, isoluminant colour samples under a variable illuminant against the standard (C) revealed variations of colour constancy and brightness independent of luminance. Constancy was found worse for illuminants below the Planckian curve (red, purple, violet), but rather similar under other illuminants. Brightness induction effects occurred over similar test hue ranges: 5PB–5GY for A, red, purple illuminants, and over 5GY–5RP hues for S and green illuminants, suggesting the operation of one dominant red-green opponent mechanism. Thus, the cardinal tritanopic axis bisects the Uniform Chromaticity plane at about 5GY outlining the border between these induction effects. Similar effects were evident for tritanopic violet, but not for yellow.

Lightness constancy and its link with cone contrast

The link between chromatic constancy (compensation for hue and saturation shifts) and lightness constancy (compensation for a change in surface reflectance) was tested theoretically by computing cone contrasts and by asymmetric matching experiments. The effect of a thin achromatic line (a frame) around the test sample was tested empirically. When the samples were outlined by the frame, lightness constancy was increased and chromatic constancy reduced (p<0.001). Changes in luminance are more likely to be compensated when the luminance contrast edge around the test stimulus is disturbed as with the addition of an achromatic frame.

Colour matching of isoluminant samples and backgrounds: A dimming effect

Sequential asymmetrical colour matching of forty Munsell samples simulated under illuminant C and one of eight test illuminants was carried out. The subjects matched the appearance of each sample under illuminant C with its appearance under the test illuminant. Samples and background (N7) were presented for 1 s under the test illuminant and were isoluminant with each other. Subjects adjusted hue, chroma, and value under illuminant C. The experiments distinguished two groups of subjects; some observers needed to reduce the luminance of the sample to make a match while others did not. This ‘dimming’ occurred when the matches were close to cardinal axes, especially the tritanopic confusion line. A model of luminance and cone-opponent mechanisms contributing to brightness can account for the dimming effect. Details of analysis in cone-opponent space (L-M, L+M-S, L+M) are presented in the companion paper (Stanikunas et al, 2005 Perception 34 this issue).

Influences of prolonged viewing of tilted lines on perceived line orientation: the normalization and tilt after-effect

Gibson [J. Exp. Psychol. 16, 1 (1993)] observed that during prolonged viewing, a line perceptually rotates toward the nearest vertical or horizontal meridian (the normalization effect), and moreover, the perceived orientation of a subsequently presented line depends on the orientation of the adapting one (the tilt after-effect). The mechanisms of both phenomena remain poorly understood. According to our experimental results, the adapting line perceptually rotates to the nearest of three orientations: vertical, horizontal, and diagonal. We propose a simple neuronal model of orientation detectors whose responses are determined by the cardinal detectors. It is shown that both normalization and tilt after-effect may be explained by adaptation of these cardinal detectors.