Chapter 14

Subjective Preferences in Vision

In order to rationally design visual objects and spaces with both temporal and spatial sensations in mind (“temporal design”), we want to understand subjective preferences for visual stimuli. A series of experiments was carried out to probe the subjective preferences for temporal and spatial factors in vision. This included preferences for rates of flickering lights, for frequencies of oscillating vertical and horizontal movements, and for texture-related spatial regularities. The temporal and spatial factors are extracted respectively from temjporal and spatial autocorrelation functions (ACFs).

14.1 Subjective Preferences for Flickering Lights

Subjective preference for a flickering light obtained using paired comparison tests (PCT) was investigated in terms of the ACF factors of the signal in the time domain. It has been found that the preferred period of the flickering light with a sinusoidal signal is approximately 1 Hz (Soeta et al., 2002a). The purpose of this study was to find more preferred conditions introducing a fluctuation to the flickering light. For this purpose, the bandwidth of the noise centered on 1 Hz was varied at five levels (1, 2, 4, 8, and 16 Hz) by use of the second-order Chebyshev filter. A remarkable finding is that the most preferred value of [φ1]p is about 0.46, where φ1 can be extracted from the temporal ACF of the stimulus signal. The scale value of preference is formulated approximately in terms of the 3/2 power of the normalized φ1 of a flickering light by the most preferred value, [φ1]p. This result may suggest a reason, for example, why one likes the twinkling star. Further, we may produce a visual light and even a music signal based on this temporal factor, and also blending visual light and music.

In the natural environment, for example, we have many visual aspects in the temporal fluctuation, such as leaves in the wind and clouds in the sky, twinking stars due to air currents, flames, and flows of water in a river. Flames in a bonfire and a glitter of sunlight reflected by the water surface provide us with a lively and splendid environment.

Although a number of studies have dealt with sensitivity to flickering stimuli (e.g., de Lange, 1952; Kelly, 1961; Mandler and Makous, 1984; Kremers et al.,

1993; Wu et al., 1996), this section is concerned with subjective preference in timevarying light. Subjective preference was initially chosen as a primitive and overall response relating to aesthetics. It would lead the individual away from inappropriate environments and toward desirable ones (Schroeder et al., 1974; Kaplan, 1987).

It has been found that the preferred period of a sinusoidally-flickering light is approximately 1 Hz. To obtain more preferred conditions, we introduced a fluctuation to the sinusoidal flickering light centered on 1 Hz (Soeta et al., 2005). The amplitude of the first maximum peak of the ACF, φ1, was controlled by changing the bandwidth of the noise (1, 2, 4, 8, and 16 Hz). This temporal factor that reflects flicker regularity is the visual analogue of auditory pitch strength. An 8-mm- diameter green LED, set at a distance of 1.0 m from the subject in dark surroundings, produced the light source. The stimulus field from the LED was spatially uniform, and its size corresponded with 0.46◦ of the visual angle. The mean luminance was set to 30 cd/m2 and kept constant during the sessions. Examples of the stimulus signal are shown in Fig. 14.1.

Fig. 14.1 Examples of stimuli used. (a) Sinusoidal wave. (b) f = 1 Hz. (c) f = 4 Hz

The source signal was characterized by the ACF factors, (0), φ1, τ1, and τe (Fig. 14.2, 2T = 4.0 s). The value of (0), representing average power, was set constant. The τ1 corresponding to the center frequency of the band-pass noise was fixed at 1.0 s (Fig. 14.3a). Note that the factor Wφ(0) is constant also. The values of φ1 and τe increase as the bandwidth of noise, f, decreases (Fig. 14.3b and c). Because

Fig. 14.2 An example of the normalized autocorrelation function NACF of a signal showing definitions of dominant periodicity τ 1 and its relative magnitude φ1. For acoustic signals these features correspond respectively to pitch and pitch strength or salience

there is a certain degree of coherence between φ1 and τe, subjective preference for a flickering light is discussed based only on φ1 in this study.

Ten 20to 23-year-old subjects participated in the experiment. All had normal or corrected-to-normal vision. They adapted to the dark and watched the LED stimulus. Then, the PCT was performed for all combinations of the pairs (i.e., 15 pairs of stimuli interchanging the order in each pair per session), and the pairs were presented in random order. A total of 10 sessions were conducted for each individual subject. The subjects were asked which stimulus they preferred to watch.

The most preferred, [φ1]p, for each subject was obtained by fitting a suitable polynomial curve to a graph on which scale values were plotted. Figure 14.4 shows an example of the method used for estimating [φ1]p. The peak of this curve denotes the subject’s most preferred value. Table 14.1 shows the results of [φ1]p for the 10 subjects. The averaged value of [φ1]p was 0.46.

We also attempted to determine the characteristics of the preference evaluation curve in more detail, in a similar manner to those described in Sections 3.2 and 7.3.

where α and β are the weighting coefficients, and x = logφ1 – log[φ1]p. After obtaining the most preferred value [φ1]p for each subject, values of α and β were obtained. To simplify Equation (14.1), the coefficient β may be fixed at a certain value so that the individual preference curve can be expressed by the sole coefficient α. The average value of β, estimated by a quasi-Newton numerical method, was approximately 1.47, as shown in Table 14.2. As discussed later, the scale value of preference for the period of a flickering light and the circular stimulus moving in vertical and horizontal directions is also formulated approximately in terms of the 3/2 power of the normalized period. Thus, the value of β was fixed at 3/2 here again, so that the coefficient α can represent individual difference in subjective preference. The

c

Fig. 14.3 Measured factors extracted from the temporal ACF of the visual source signal as a function of the bandwidth f. (a) Delay time of the first maximum peak of ACF (τ1). (b) Amplitude

of the first maximum peak of ACF (φ1). (c) Effective duration of ACF (τe). The τe of the pure tone is ∞

values obtained by a quasi-Newton numerical method are listed in Table 14.3. The weighting coefficient α describes the sharpness of the preference curve with respect to φ1. The large α value signifies that the subject clearly differentiates the level of preference.

Figure 14.5 shows scale values for all of the subjects and the preference evaluation curve calculated by Equation (14.1). The correlation coefficient between scale values of preference and calculated values by Equation (14.1) is 0.92 (p < 0.01). Remarkably, the resulting preferred fluctuation expressed by φ1 is an intermediate of the values between those of sine and a perfectly random wave.

The scale value of preference is formulated commonly in terms of the 3/2 power of x of a flickering light. This behavior is consistent with preference judgment for a flickering light without fluctuation (Soeta et al., 2001); a circular target