溝上 陽子

本研究では、安定した色知覚のためには空間・照明認識、環境のナチュラルネスが大きく関わることを示すため、視空間のナチュラルネスが色の恒常性に与える影響を検討する。空間の構成を不自然にして通常の空間・照明認識が出来ない場合、色の恒常性の成立度が低くなることを示す。 実験ブースは、窓を介して奥行き方向に並んだ2つの小部屋と被験者室で構成され。奥の部屋（後室）は昼白色、手前（前室）は電球色で照明された。被験者は観察窓を通して、室内と各室中央に置かれた色判定用のテストパッチを観察した。観察条件は自然条件と、万華鏡形式の観察ボックスを通して部屋を見る不自然条件の2種類である。不自然条件においては、周辺視野の配置は置き換わるが、視野中央のテストパッチと隣接する背景は自然条件と全く同じである。被験者は黒体軌跡上を段階的に変化するテストパッチに含まれる色みをRGBYを用いて答えた。色系列のうちRYとBGの判定が拮抗する点をニュートラル知覚とした。 結果は、前室において不自然条件では自然条件と比較してニュートラル知覚が昼白色方向にシフトした。すなわち色の恒常性の成立度が低くなった。これにより視空間のナチュラルネスが色恒常性に影響を与えることが確かめられた。

Wavelength-selective filtering by the lens and macular pigment biases the spectra of broadband lights more than narrowband. We asked to what extent these filtering effects influence the perceived hue of a light as its bandwidth is varied. Previously we approached this question by comparing individual differences in unique hues as a function of bandwidth (Mizokami et al., JOV 2002). In the present study we instead sampled a range of spectral stimuli with a hue-matching task. Spectra were created by dispersing light with an interference wedge and then filtering by spatial masks on an interposed LCD panel (Bonnardel et al., Vision Research 1996). The lights were displayed as a uniform 2-deg field formed by the aperture of an integrating sphere. Subjects viewed the field directly or at an eccentricity of 10 deg. Reference stimuli had a fixed dominant wavelength and a roughly Gaussian spectrum with a bandwidth of 105 nm. Test stimuli had a bandwidth of 25 nm or greater. On each trial the two lights were shown in succession. The dominant wavelength of the narrower bandwidth was varied in a 2AFC staircase to match the pair for hue. We compare these matches to the predicted spectra required to maintain a fixed direction in cone-excitation space. Results suggest a weak but incomplete influence of pigment screening at shorter wavelengths, suggesting partial compensation for the filtering effects of the eye.

The distribution of colors in natural images can vary widely across different environments or within the same environment over time. We have examined how color distributions at a single location can vary with changes in seasons. A large set of digital images were collected from a rural valley in Maharashtra India during monsoon (wet) and winter (dry) months. Each image included a reference palette of known reflectance (MacBeth color checker measured with a spectroradiometer), which allowed the rgb values to be calibrated to derive the cone excitations at each pixel. The seasonal changes in rainfall and consequent changes in vegetation result in large changes in both the mean chromaticity of the images and the dominant chromatic axis along which color signals vary. Mean color shifts were largely along the L-M chromatic axis (shifting toward +L for drier environments). This average seasonal shift is notably similar to the +L shift that allows ripening fruit to be distinguished from the background foliage (Regan et al., Phil. Trans. R. Soc., 2001). The principal axes of the distributions rotated from near the S-LM axis to a bluish-yellowish axis intermediate to the L-M and S-LM axes. These shifts are consistent with previous measurements sampled from different environments (Webster and Mollon, Vision Research, 1997). Both the mean changes and the axis changes are many times larger than the differences resulting from variations in the illumination, and have important implications for understanding how visual coding might be matched to particular environments.

Laboratory studies with controlled stimuli have shown that adaptation to natural variations in faces can strongly bias natural categorical judgments of faces (e.g. of gender, ethnicity, or expression). We tested for signs of these adaptive adjustments outside the lab, to examine how adaptation might influence face perception in "real" natural environments. To assess this, we probed judgments of ethnic category (Caucasian vs. Japanese) in observers who were exposed to different face populations. Participants were newly arrived (< 1 week) exchange students from Japan and Caucasian students in Reno, NV. Stimuli were morphs between Japanese and Caucasian face images from the Matsumoto and Ekman neutral-expression set. Observers made forced-choice judgments of ethnicity, with the morph level varied in a staircase to determine their category boundary. Japanese and Caucasian students chose boundaries closer to their own categories, suggesting that observers are more sensitive to how stimuli differ from their individual category. To test whether these boundaries might adjust in a changed environment, we tested a second group of Japanese students who had been resident in the US from 18 to 72 months. The mean boundary for this group was intermediate to and significantly different from both the Caucasian and newly-arrived students. Individual boundaries were also positively correlated with their self-reports of time in the US and negatively correlated with their report of the percent of time they spent with their own ethnic group. These shifts are consistent with an adaptive change in face coding that renormalizes face perception according to the average set of faces observers are exposed to.

A central question in color vision is: to what extent are the unique hues tied to properties of the environment versus properties of observers? We are exploring this by asking how the unique hues vary with changes in the spectral bandwidth of the stimuli, for which environmental and physiological accounts make different predictions. Selective filtering by the lens and macular pigment differentially biases the spectrum of broader bandwidth lights relative to narrower bandwidths. We are examining the extent to which these biases influence differences in perceived hue both across individuals (who might vary in ocular pigment density) and within the same individuals at different retinal eccentricities (which differ in macular pigment density). Stimuli were roughly Gaussian spectral distributions, formed by passing light through an interference wedge masked by spatial patterns generated on an LCD panel (Bonnardel et al. Vision Research 1996). Bandwidths ranged from ∼20 nm to 80 nm (full width at half height) and were empirically matched for brightness with an achromatic background (1.7 cd/m2). Observers made unique blue, green, and yellow settings in the fovea or at 10 deg in the periphery. In a second task we also made hue matches between lights of different bandwidth. Unique hues within observers shift toward shorter wavelengths in the periphery, consistent with previous studies and in rough agreement with the differences predicted by variations in macular pigment density. Differences across observers are compared to two possible models of the unique hues (based on assuming a common set of cone ratios or a common stimulus in the environment) which differ in predicting either a decrease or increase in interobserver differences as the spectrum narrows.

We recognize the outside world as a 3-D space in spite of its two-dimensional retinal image. We demonstrated a two-dimensional photograph could be perceived as a 3-D scene in a special 'dimension-up' viewing condition that a subject observed only the photograph. The color constancy was then realized in part even in the photograph and its degree increases depending on the degree of 3-D recognition. A jumbled photograph was made from an original photograph taken for a room under incandescent lamps. Either was on a wall of an experimental booth illuminated by white light. In the normal viewing condition, the subjects perceived neutral white for almost the same test stimulus whether in the original or the jumbled. In the dimension-up viewing condition, the shift of the neutral perception for the original photograph was larger than for the jumbled. This should indicate that the recognized visual space of illumination (RVSI) for the scene illuminated by incandescent lamps was constructed for the original photograph and the test stimulus was perceived as an object in the scene. The degree of the color constancy was larger in the photograph perceived as a 3-D scene than in that perceived as a mere two-dimensional scene.