Key Points
We estimate a body shape model from a single depth image and select a clothing image captured from a person with a similar body shape. We then transform the selected clothing image and overlay it on the user’s image to create a natural virtual fitting experience.
Abstract
This study aims to let people try on clothes on a screen without changing what they are wearing. Simply placing a clothing photograph over a person's video image can look unnatural if their body shapes do not match. We therefore propose selecting clothing images based on the user's body shape.
The system uses a depth image, which records the distance from the camera to the body. From a single capture, it estimates the body's three-dimensional shape and selects a clothing image photographed on a person with a similar shape. It then transforms that image to fit the user and overlays it on the user's video in real time.
Our observations showed that clothing images from people with similar body shapes fit the user more accurately. Based on this finding, we developed a fast way to estimate body shape and select images. We demonstrated several virtual fittings and evaluated how well the clothing images fitted, comparing the method with an existing approach that does not consider body shape.
Publication
- Masahiro Sekine, Kaoru Sugita, Frank Perbet, Bjorn Stenger, and
Masashi Nishiyama,
Virtual Fitting by Single-shot Body Shape Estimation, Proceedings of 3D Body Scanning Technologies, pp. 406 - 413, October 2014.
Publication (Japanese)
- 杉田 馨, 関根 真弘, 西山 正志,
人物シルエットへの時空間位置合わせによる衣服画像の合成,
電子情報通信学会論文誌 D, Vol.J97-D No.9, pp.1519-1527, 2014.
Short movies
