Estimation of impression of store interior design based on color features extracted using object segmentation

Abstract

The aim of this study is to analyze the impression provided by a color image and the color information contained in the image. To process a large number of color images, we used a method of extracting representative colors from an image based on pixel information. In this study, we developed a method to combine image recognition technology using deep learning with representative color extraction technology, and analyzed the impression and color characteristics of design for store interior images. We collected color images of store interiors along with their tagged keywords from the web. Representative colors were extracted via a method using region division by deep learning. Furthermore, we analyzed the characteristics of the representative colors included in the corresponding images for the keywords "natural," "modern," and "cute," which were frequently used in the collected data to express the impression of the store design. The tendency of color features was analyzed for each image. For example, in "natural" designs, there were many colors close to yellowish green, which were associated with objects such as "plant" and "tree". Such knowledge is important for design support, that is, knowledge of the color to be incorporated into the design, as well as of methods to incorporate that color.