Texture Overlay onto Flexible Object with PCA of Silhouettes and K-means Method for Search into Database
Hiroshi Tanaka, Hideo Saitô · Machine Vision Applications · 2009
In this paper, we propose a method to overlay an arbitrary texture onto a T-shirt worn by a user. To realize such a system, two phases, an offline phase and an online phase are required. During the offline phase, we use a T-shirt with several markers for training images and establishing a correspondence between the shape of the T-shirt area and the array of markers. During the online phase, we use a T-shirt without any markers for the input image and search for a training image which has the most similar shape to the input image. By using markers from the selected training image, we overlay a texture onto the input image. In this method, we represent the shape of a T-shirt based on its contour so that we can handle occlusion by the hands when the hands cross over the T-shirt. In addition, the computation time is reduced by using PCA and K-means method, which enables our system to work in real time.