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

The fringe projection profilometry (FPP) technique has always been the focus of research and attention from numerous scholars. However, when obtaining the fringe pattern, the light of the image may be too low due to environmental factors, which may affect the smooth progress of subsequent work. To address this issue, this paper proposes an LFE-Net network for enhancing low-light fringe patterns. In this method, RGB fringe patterns and grayscale fringe patterns are used as the input feature map of the entire network. The wavelet transformation preprocessing (WTP) module is designed to process the input feature map and obtain rich feature information. At the same time, the residual convolution transformer parallel (RCTP) module is designed to ensure the network model's ability to connect contextual information and enhance the flexibility of the network. The experimental results on our dataset show that our proposed method outperforms the multi-scale retinex (MSRCR) algorithm, MIRNet, and HWMNet in processing low-light fringe patterns. Moreover, under the same hardware environment, the computational time of the proposed method is only 8.66% of MIRNet and 40.49% of HWMNet.