Impression estimation model and pattern search system based on style features and Kansei metric

Abstract

In this study, we aimed to construct impression estimation models of clothing patterns based on style features and Kansei metric. We first conducted a subjective evaluation experiment and a factor analysis, and quantified visual impressions of flower patterns. Following that, we used style features using CNN as image features suitable for representing flower patterns. Then, with a Lasso regression, we reduced the dimension based on Kansei metric (impression evaluation) and modeled the relationship between visual impressions and image features. Furthermore, we implemented a pattern search system using the modeled relationship.