Fig. 1. Optical path schematic of traditional Zernike phase contrast microscopic imaging system^[4]

Fig. 2. Flow charts of network training and testing. (a) Flow chart of data collection and network training; (b) flow chart of network testing

Fig. 3. Partial data sets. (a) Defocused microscopic images of red blood cells; (b) defocused microscopic images of onion epidermal cells; (c) defocused microscopic images of human skin perspiration glands slices

Fig. 4. Neural network model. (a) DLFP network model structure; (b) operation diagram of Gather-Excite module，where H is the height, W is the width, C is the number of channels, W′ is the rounded up (W/e), H′ is the rounded up (H/e), and e represents the ratio of selected ranges

Fig. 5. Discriminator network structure

Fig. 6. Test results. (a)(d)(g) Network input microscopic images; (b)(e)(h) Ground-truth; (c)(f)(i) output results of DLFP network

Fig. 7. Comparison of pixel values between Ground-truth and center line of DLFP network output results. (a) Red blood cells; (b) onion epidermal cells; (c) human skin perspiration glands slice

Fig. 8. Output results of different network models. (a) Red blood cells; (b) onion epidermal cells; (c) human skin perspiration gland slice

Fig. 9. Test results of a small number of data sets in generating adversarial network. (a) Test result of red blood cell sample; (b) test result of onion epidermal cell sample; (c) test result of human skin perspiration gland slice sample