The Basic Study of the Impression Word Extraction from the Object Appearance using CLIP

Abstract

Humans are creatures that place great importance on visual information, and form impressions of objects based on a combination of information such as shape, color, and texture, as well as on their identification. These visual impressions are also important in marketing and product development. In the field of Kansei informatics, impressions of such objects are expressed using adjectives in questionnaire surveys using the Likert scale or the Semantic Differential (SD) method. However, when actually conducting a questionnaire survey on a product, it is necessary to manually select a limited number of adjectives. In this study, due to automatic extract adjectives, CLIP is validated for its ability to convert images directly into adjectives.