文献类型：期刊文献

英文题名：Self-adaptive and efficient image restoration based on stripe geometric modulation for three-dimensional reconstruction of high-dynamic-range surface

作者：Yin, Xiaoqia[1];Cheng, Huayi[1];Kong, Dechang[1];Cheng, Haonan[1];Zhang, Xiancheng[1];Tu, Shandong[1]

机构：[1]East China Univ Sci & Technol, Minist Educ, Key Lab Safety Sci Pressurized Syst, Shanghai, Peoples R China

年份：2025

卷号：64

期号：8

外文期刊名：OPTICAL ENGINEERING

收录：;EI(收录号：20253719140579);WOS:【SCI-EXPANDED(收录号:WOS:001565529800022)】；

基金：This work was financially supported by the National Natural Science Foundation of China (Grant No. 52005333), the National Defense Basic Scientific Research Major Project of China (Grant No. JCKY2023203A002) and the Shanghai Gaofeng Project for University Academic Program Development.

语种：英文

外文关键词：high-dynamic-range surface; stripe geometric modulation; structured-light projection; image restoration

摘要：Surface three-dimensional (3D) reconstruction is common in engineering applications. The stripe scanning technique based on the optical triangle principle plays a crucial role in the surface 3D reconstruction due to its advantages of non-contact, high performance, and low cost. However, high-dynamic-range (HDR) surfaces present a huge challenge to the stripe scanning technique. The mainstream solutions always project several sets of stripe patterns under different exposure conditions to fuse an ideal image, which need resetting the exposure conditions when the HDR surface changes, resulting in inefficiency and non-intelligence. We proposed a self-adaptive and efficient image restoration method for the HDR surfaces, where a single-shot picture is enough and recalibration for different surfaces is unnecessary. The proposed method mainly includes two parts: first, stripe edges around the HDR region are extracted by a specially proposed extraction method; second, the HDR regions are restored based on the stripe geometric modulation. Finally, 3D reconstruction experiments for HDR surfaces are implemented to verify the feasibility of the proposed method. Moreover, the proposed method is discussed in application scope, accuracy, efficiency, self-adaptivity, and hardware requirements, which shows that the proposed method is applicable to fluent surfaces but not delicate surfaces; compared with the exiting methods based on images synthesis, the proposed method needs only one image, no recalibration and can reach a close accuracy; compared with the exiting methods based on hardware assistant, the proposed method can improve accuracy significantly by 87%.