OPTICS & OPTOELECTRONIC TECHNOLOGY, Volume. 23, Issue 2, 93(2025)
Neural-Radiance-Fields-Based Framework for Novel View Fringe-Pattern Phase Synthesis
In the research field of fringe-pattern phase analysis,the use of digital simulation and modeling tools for simulation experiments can improve the efficiency and convenience of research. However,due to the limitations of simulation algorithms and modeling tools,the data generated by simulation experiments often deviate significantly from the actual situation. To address this issue,a framework based on neural radiance fields(NeRF)is proposed. After training on a specific three-dimensional scene,this framework can render accurate fringe-pattern phase images for any given pose. Experimental results show that the mean squared error(MSE)of the phase within the region of interest(ROI)of the newly generated view images remains at the order of 10-5. When the dataset generated by this framework is used for deep learning training,it can achieve model accuracy at the same order of magnitude as that of a real dataset. The proposed framework can effectively generate novel-view fringe-pattern phase images for use in deep learning training.
Get Citation
Copy Citation Text
RAN Chen-xun, XIN Jing, ZHANG Qi-can, WANG Ya-jun. Neural-Radiance-Fields-Based Framework for Novel View Fringe-Pattern Phase Synthesis[J]. OPTICS & OPTOELECTRONIC TECHNOLOGY, 2025, 23(2): 93