Object Image Relighting through Patch Match Warping and Color Transfer

Xin Jin1, *, Yulu Tian1 , Ningning Liu3 , Chaochen Ye1 , Jingying Chi2 , Xiaodong Li1, *, Geng Zhao1

1Beijing Electronic Science and Technology Institute, Beijing 100070, China

2 Beijing University of Chemical and Technology, Beijing 100029, China

3 University of International Business and Economics, School of Information technology and Management, Beijing, 100029, China * Corresponding Authors: {jinxin, lxd}@besti.edu.cn

Abstract

      In this paper, we propose a novel method for image based object relighting through patch match warping and color transfer. According to our method, a pair of reference object images (A and B), without any knowledge of the 3D geometry or material information of the input object image, are needed. The input image has similar illumination condition as that of the reference image A. The reference image B has the target illumination condition. We first modify the color of the reference object images according to the color of the input object image. Then the modified reference image A is warped to the input image using the image patch match. The reference B has the same warping to that of A. At last, the illumination of the reference B is transferred to the input image through a local and global transfer model learned from the warped reference B and A. We test our method on the ALOI dataset (Amsterdam Library of Object Images) and acquire convincing relit result on numerous input and reference object images with different illumination conditions.

Gallery

icvrv2016

Relighting with and without color modifying. (a) is the input image. (b) and (c) are the original reference image A and B. (d) is the relit result directly using (b) and (c) without color modifying. (f) and (g) are the color modified version of (b) and (c). (h) is the relit result using (f) and (g). (e) is the ground truth.

Paper and Slides

pdf

Xin Jin, Yulu Tian, Chaochen Ye, Jingying Chi, Xiaodong Li*, Geng Zhao.The 16th International Conference on Virtual Reality and Visualization (ICVRV), Hangzhou, China, 23-25 September, pp.235-241, 2016.

video

mp4

发表评论

电子邮件地址不会被公开。 必填项已用*标注