Impression Estimation Model for Clothing Patterns Using Neural Style Features

Abstract

In product design, the impressions evoked by surface properties of objects attract attention. These impressions are called affective texture and demand is growing for technologies that quantify, index, and model it. Even in the fashion industry, the diversification of user needs necessitates the customization and personalization of products. Consequently, there has been a focus on custom-made services. However, enormous amounts of time and human costs are needed to find clothing patterns to suit one’s own preferences and ideas from among the countless patterns available. This research focused on affective texture related to visual impressions, and we proposed a method for automatically estimating the affective texture evoked by clothing patterns. To this end, we conducted the following steps: (1) quantified the visual impressions for patterns; (2) extracted style features as physical characteristics; and (3) modeled the relationships between visual impressions and physical characteristics. Afterward, based on the obtained models, we estimated the impressions for unlabeled patterns. Then, we verified their validity through relative evaluation and absolute evaluation, and we confirmed that our models estimated the impressions corresponded to the impressions that people actually felt. In addition, we implemented a system to enable users to intuitively search for patterns.