364 JOURNAL OF COSMETIC SCIENCE Frontal facial images of 243 Japanese women aged 20–75 years were used. Written informed consent was obtained from all subjects, allowing the use of photographs for research studies. Facial images were taken in a light box using a white light-diffusion fabric. The subjects placed their faces on a chin rest placed in the light box for photographs. Six strobe lights were set on the left and right front sides (three lights for each) of the light box to illuminate the subject’s face through the diffusion fabric. A single-lens reflex camera (Nikon D300S) with a 105 mm lens Micro-NIKKOR (Nikon, Tokyo, Japan) was placed 1.5 m in front of the subject. The exposure time was 1/250 seconds, f-stop value was 7.1, and ISO was 200. The captured image size was 2,842×4,288 pixels, which was saved as a raw image (Nikon Electronic Format). For all facial images, the RGB values of each pixel were converted to L*a*b* with color calibration using the RGB values of a color chart (X-Rite ColorChecker Mini Chart, X-Rite, Grand Rapids, MI, USA) installed on their chest. From the L*a*b* values of the images, the sclera of the eye was detected using a face detection library, ASMLibrary, and by setting a threshold for L*a*b* values in MATLAB 2016a (MathWorks, Natick, MA, USA) (15). The L*a*b* value of the detected sclera area was measured for all facial images, and its correlation with age was evaluated. PSYCHOPHYSICAL EXPERIMENTAL METHOD In this study, the perceived sclera color and eye size of facial images were measured using various degrees of eyeliners. If eyeliners affect perceptual sclera color and eye size, the measured perceptual color and size should differ from their actual values. The staircase method was employed to measure the point of subjective equality (PSE) as the perceptual sclera color and eye size (16). This experimental method used two groups of stimuli: standard and comparison. The standard stimuli in this study were the measurement targets, eyeliner-applied images. In contrast, the scale images of perception were used as comparison stimuli. In this case, two image datasets given by sequentially modulated sclera color and eye size were prepared as comparison stimuli sets. The PSE for each standard stimulus was measured by determining a perceptually equivalent point on the scale of the comparison stimuli. STIMULUS IMAGES An average face image, generated from the images of 40 Japanese females aged 20–26 years, shown in Figure 1, was used as the stimulus image. Image processing to create standard and comparison stimuli was performed using ASMLibrary and imwarp function in the Image Processing Toolbox of MATLAB 2016a. Eyeliner-drawn images were prepared as the standard stimuli, while the eye-size–modulated images and sclera-color–modulated images were used as the comparison stimuli for the psychophysical experiment. Three stimuli sets were formed from the eyeliner-drawn images: 1) black eyeliners with thickness modulation 2) hue-rotated colored eyeliners and an equal-lightness achromatic eyeliner and 3) brown and blue eyeliners with two different light levels, respectively. Stimuli sets 1, 2, and 3 were used in Experiments 1, 2, and 3, respectively. Experiment 1 evaluated the influence of black eyeliner on perceptual sclera color and eye size, and their

365 EFFECT OF EYELINER ON THE PERCEIVED SCLERA COLOR relationship with eyeliner thickness. Experiment 2 evaluated the relationship between the hue of eyeliner color and its effect on perception. Experiment 3 evaluated the effects of lightness and hue of eyeliner on perception, color of the sclera, and eye size. The image processing for the thickness-modulated stimuli is presented in Figure 2. First, the outline of the applied eyeliner was manually defined, and the average vertical line width from head to tail was computed. The line width was modulated according to the target thickness (0.5 to 2.0 mm with 0.5 mm step size), and the inside of the outline was filled with black color (R =G =B =0.1). The obtained black eyeliner layer image was placed on the original face image using alpha blending (blending ratio of 0.9). The colored eyeliner images for stimuli sets 2 and 3 were generated following the same procedure by Figure 1. Average face image of 40 Japanese females aged 20–26 years. Figure 2. Image processing steps for virtual eyeliner application.