Figure 3
From: Deep learning alignment of bidirectional raster scanning in high speed photoacoustic microscopy
Performance comparison of the MS-FD-U-Net with other existing methods. (a) OR-PAM images reconstructed with bidirectionally (marked as Input) and unidirectionally (marked as Ground truth) acquired data, the MS-FD-U-Net GAN, an upsampling method (i.e., bicubic interpolation), conventional filtering methods (i.e., bilateral and median filtering), and other DNNs (i.e., Dense GAN and FD-U-Net). PA amplitude profiles from the regions highlighted by the (b) blue and (c) green dashed lines in (a), respectively. The graphs display the profiles in the images of Input, Ground truth, MS-FD-U-Net GAN, FD-U-Net and median filtering. The graphs for other methods (i.e., bicubic, bilateral filtering and dense GAN) are displayed in Supplementary Fig. S4.
