LFE-Net: a low-light fringe pattern enhancement method based on convolutional neural networks.

Area of Science:

• Optical Metrology
• Computer Vision
• Image Processing

Background:

• Fringe Projection Profilometry (FPP) is crucial for 3D measurements.
• Low-light conditions degrade fringe pattern quality, hindering accuracy.
• Existing enhancement methods struggle with low-light FPP images.

Purpose of the Study:

• To develop an effective deep learning network for enhancing low-light fringe patterns.
• To improve the robustness and accuracy of FPP under adverse lighting.
• To introduce a computationally efficient solution for low-light image enhancement.

Main Methods:

• Proposed LFE-Net architecture utilizing RGB and grayscale fringe patterns.
• Incorporated Wavelet Transformation Preprocessing (WTP) for rich feature extraction.
• Employed Residual Convolution Transformer Parallel (RCTP) for contextual information and flexibility.

Main Results:

• LFE-Net demonstrated superior performance in enhancing low-light fringe patterns compared to MSRCR, MIRNet, and HWMNet.
• Achieved significantly faster computation times: 8.66% of MIRNet and 40.49% of HWMNet.
• Experimental results validated the effectiveness and efficiency of the proposed network.

Conclusions:

• LFE-Net offers a robust solution for low-light fringe pattern enhancement in FPP.
• The network provides high-quality results with improved computational efficiency.
• This work advances the application of FPP in challenging lighting environments.