抄録
An image emotion space defining the relationship between the manipulation of a photographic image and the resulting change in the impression it produces was constructed by presenting original images and their manipulated versions in pairs to a panel of ob- servers and asking them to identify the one giving the stronger impression for each of 59 emotion words. Picture-independent results could be obtained by comparing raw images of various pictures with those having various effects applied at random. Components governing the responses were extracted from the results by principal component analysis to construct a two-dimensional space with the components lightness and contrast and chroma as axes. Furthermore, it was shown that the differences in impressions can be clustered into 12 clusters by cluster analysis. A Kansei space model showing pathways to “preferred” was obtained by multiple linear regression analysis using “preferred” as the dependent variable and the representative words of their respective clusters as the inde- pendent variable. It consisted of two pathways, one for lightness information and one for color information.
https://ror.org/02h4myp42 
