Image-based virtual try-on system with clothing extraction module that adapts to any posture

Abstract

In this study, we developed a virtual try-on system that automatically creates a clothing model from a person’s image, and thus a clothing model is not required as an input. Recently, the scale of the E-commerce market in the fashion field has expanded. When purchasing clothes on a fashion E-commerce site, a user can browse and buy various brands at once without visiting a physical store. However, users cannot pick up the actual clothes and try them on. Therefore, the user may not be satisfied with the design and size after receiving clothing bought on-line. Virtual try-on systems have been developed in various studies that allow users to virtually try on clothes using a computer. In particular, many image-based virtual try-on systems have been reported that transition the clothing model’s features to the clothing area on the target person. However, this method has the following problems: (1) it is necessary to prepare a clothing model in advance and (2) the posture for changing clothes is restricted. Thus, we developed a virtual try-on system that does not require a clothing model as an input, thereby enabling the automatic creation of a clothing model from a person’s image by incorporating a clothing extraction module into the virtual try-on system. In addition, we improved the virtual try-on system so it can adapt to any posture and evaluated the improvements in simulations.