Smoothness and flicker perception of temporal color transitions

We present results from two experiments designed to explore temporal properties of human color vision relevant to dynamic lighting applications. Sensitivity for smoothness perception of linear temporal transitions and flicker visibility was tested. Stimuli in the first experiment were linear color transitions, varying in either lightness, chroma or hue, around a base color represented in CIE LCh. Results show a significantly lower smoothness threshold for lightness changes than for chroma and hue changes. Moreover, the thresholds for lightness change show independence from the chroma and hue of the base color in contrast to thresholds for chroma and hue changes. A difference between the sensitivity for chroma and hue changes was also demonstrated. In the second experiment, the sensitivity for linear transitions is compared to flicker sensitivity for the same base colors. Results show that visibility thresholds for flicker are significantly lower than the thresholds smoothness of linear changes, demonstrating an influence of the type of change to the temporal sensitivity. The results from the flicker experiment show the same tendencies as the linear changes. The results from these experiments show a need for a model of perceived smoothness to control temporal changes in dynamic lighting systems and give the first steps towards building such a model. Introduction Advances in lighting, especially in Solid State Lighting, enable new uses of light. Having improved spatial and temporal resolution, more saturated primaries and lower power consumption, LED based lighting systems can be used to design more complex and attractive lighting atmospheres. One of the largest differentiators of such lighting systems are their dynamic capabilities. However, the produced dynamic lighting atmospheres need certain properties to be attractive to the users. Perceived smoothness is one of those desirable properties. Aside from a limited set of applications, such as for disco lights and concerts, or being used as attention attractors, abrupt changes in environment lighting are hardly perceived as pleasant. Other properties of the produced lighting atmospheres that might have an influence on the attractiveness such as the hue composition or the spatial distribution, moreover, are more subjective. The work presented in this paper studies temporal properties of human color vision relevant to dynamic lighting applications, namely thresholds for smoothness perception of linear transitions and flicker visibility. To understand the possible source of problems connected to smoothness perception, we discuss the design of modern lighting systems and applications first. Modern lighting applications use discrete control of the light sources, with a limited number of intensity levels. Contrary to the analog systems which have a continuous change in color, in digital systems the smallest distance between two colors, both in color and time, is limited by the resolution of the system. Similar to spatial color perception, an unapropriate minimum distance between colors can introduce perceived discontinuities. Existing dynamic lighting systems use the device color space (usually RGB) of the lights to control the temporal changes. To produce smooth light transitions, low pass filters are applied on the individual color channels. Under some conditions, especially for light effects computed from another medium (such as a video signal for Philips ambiLightTM), this leads to seemingly unsolvable problems. If the parameters of the low pass filter are tuned such that the transition from low intensity to high intensity of the lights appear smooth, the transitions between chromatic colors are perceived as too slow. In the case of content dependent dynamic lighting, notably for video, this introduces a mismatch between the color of lighting and the representative color of the video frames during the transition. A video transition from a red sunset to a blue underwater scene is followed by a light transition being purple for a noticeable time. This behavior is deemed undesirable by most users. The above mentioned problem is present in all dynamic lighting systems that control the temporal changes in a device color space. The core of the problem is that using a device color space, the properties of the human visual system, which determine the perceived qualities, are not taken into account. Previous work on the temporal properties of the human visual system shows differences in the way intensity and chromaticity changes are perceived. Namely, the human visual system processes intensity changes faster than chromaticity changes [1, 2]. Moreover, the changes in chromaticity are smoothed by the human visual system more than the changes in intensity [3, 4]. Using a device color space to control the temporal changes does not allow the use of such results. To compute the required distances between colors that produce spatial patterns which appear smooth, the notions of visibility threshold and just noticeable difference [5] were introduced. The continuation of the work on spatial just noticeable differences led to development of, among others, the CIE Luv, CIE Lab, and CIECAM97s color spaces [6], which show a relatively good uniformity in the predicted differences, thus predicted smoothness of spatial patterns. Unfortunately, no such spaces exist for temporal patterns. The fact that the perception of the temporal transitions depends on the frequency at which the changes are made, further complicates the representation and smoothness prediction in the temporal case. To gain better understanding of the way the human visual system processes temporal patterns in the context of dynamic lighting applications, we designed two experiments and present the results. Previous work on temporal properties of the human visual system closest to the topic of interest of this paper comes from the area of flicker sensitivity. In [7, 8], De Lange describes flicker 112 Copyright 2007 Society for Imaging Science and Technology sensitivity at different frequencies and for different average luminance levels and types of stimuli. The results of De Lange were supported by results of Kelly [9, 10], in which he additionally studies the effects of the surround average luminance on flicker perception and spatio-temporal effects. Using different methods, several authors report differences between sensitivities to luminance and chrominance flicker. In [4], the response of the visual system is modeled as a finite impulse response filter and differences in the properties of luminance and chrominance flicker were demonstrated. Kelly [11] uses spatio-temporal properties to show a difference between the luminance and chrominance flicker effects. This work differs from previous work in a number of points. First, in one of the experiments presented we use discrete linear transitions, while in prior work either flicker or a fixed number of rectangular pulses were used. Second, The chromaticity changes are further subdivided in chroma and hue changes. And third, the thresholds are computed for a number of points not only with different luminance, but also chrominance coordinates. To be able to compare the temporal sensitivity results to results from spatial visibility thresholds, the presented experiments use the CIE LCh color space and the ∆Eab metric. Based on previous studies on temporal properties of human vision described before, we form and test the validity of two main hypotheses. • There is a difference in the sensitivity for lightness and chromaticity changes. The thresholds for lightness are lower. • There is an effect of the base color point on the visibility thresholds for lightness, chroma and hue changes. Smoothness thresholds for linear color transitions The first experiment was designed to measure the sensitivity for discontinuities in linear temporal color transitions. The transitions are built in the CIE LCh color space and the threshold are expressed in ∆Eab, the Euclidian distance in the CIE Lab color space. Given the results of previous research on related topics discussed above, we investigate the effects of frequency of the changes, base color point and direction of change on the visibility threshold tested. For possible directions of change, directions parallel to the axis of the CIE LCh color space were taken.