斑点图案
计算机科学
相位恢复
光学
融合
人工智能
散斑噪声
计算机视觉
编码(社会科学)
相(物质)
相位展开
空间频率
模式识别(心理学)
散斑成像
信号处理
相位调制
编码(内存)
图像处理
电子散斑干涉技术
传感器融合
作者
Wenjie Li,Yonglong Pan,Guijie Liu,賴奇裕,Haowei Liu,Yang Huang,Zhaoguang Huang
出处
期刊:Applied optics-OT
[Optica Publishing Group]
日期:2026-10-20
卷期号:66 (16): D12-D12
摘要
To address the challenge of balancing speed and accuracy in structured light-based 3D reconstruction, this paper proposes a single-image phase retrieval method based on the fused coding of fringe patterns and digital speckle. By employing the GD-DCUNet model—integrated with multi-branch dilated convolution for feature extraction, a residual fusion mechanism, and an information integration and distribution strategy—the method achieves the effective separation of fringe pattern and digital speckle information from the fused coded image. With the assistance of a newly constructed CSAS-UNet, the separated fringe pattern image is used to predict the numerator and denominator terms of the wrapped phase. Concurrently, the separated digital speckle image is utilized to predict the fringe order via correlation calculation with a reference speckle image. Under a multi-constraint strategy involving epipolar lines, wrapped phases, and connected domains, this correlation calculation significantly narrows the search range of candidate points matching the sampling points, thereby providing robust and rapid fringe order prediction. Comparative experiments demonstrate that the proposed method achieves high-precision phase retrieval using only a single image. It maintains exceptional stability and reconstruction quality even on complex surfaces and low-texture regions, indicating promising potential for engineering applications.
科研通智能强力驱动
Strongly Powered by AbleSci AI