Fig. 22: Flowchart of computational denoising based on deep learning for phase information and phase denoising results of different methods.

a The flowchart of DnCNN-based phase denoising approach: the sine and cosine images of the noisy phase map is fed into a DnCNN to achieve the denoised phase information. To improve performance, 1–5 iterations are introduced in the denoising process. b The raw noisy phase. c The denoised phase processed with WFT¹¹⁴. d The denoised phase processed with deep learning. e The phase difference between (c) and (d). a–e Adapted with permission from ref. ³⁶², AIP Publishing