Color Appearance Evaluation of Different Light Sources by Principal Component Analysis

Abstract

The color appearance of domestic and commercial spaces is becoming increasingly important. Therefore, indices for evaluating visual impressions of lighted objects, such as vividness and brightness, as well as for measuring color preference, are needed. In this study, we conducted visual experiments to evaluate differences in color properties, such as vividness and brightness, as well as color preference, due to the spectral distribution of luminaires used in domestic and commercial spaces, and attempted to predict the experimental results using a color appearance model. Three typical LED light sources (1: blue LED+yellow phosphor; 2: blue LED+RG phosphor; 3: UV LED+RGB phosphor), a fluorescent lamp (three wavelengths) and an incandescent lamp were used as the test light sources. Visual experiments were conducted to compare the color appearance under test and reference light sources of 15 color samples, using the color rendering index. The subjects were 45 young adults and 21 elderly adults. The experimental results were evaluated by principal component analysis. Five principal components were extracted from the results of the young subjects, and four from those of the elderly subjects. In addition, the components for the young and elderly subjects differed in terms of the descriptive adjectives that had large principal component loadings; young people tended to prefer vivid and bright colors, while elderly people tended to prefer calming colors. We attempted to predict the results of principal component analysis using multiple regression analysis with the CIECAM02 model parameters. Strong correlations were obtained between the principal components representing “vivid” and “bright” and a regression equation of the lightness of high chroma colors. The determination coefficient (r2) was 0.66 for young subjects and 0.63 for elderly subjects. The results of this study indicated that predictions based on CIECAM02 parameters could be applied to the results of principal component analysis.