Fig. 1. Whole frame of TESR

Fig. 2. Edge enhancement module. (a) Four kinds of trainable Sobel operators; (b) process of our module

Fig. 3. Comparison of traditional Transformer module (left) and PLT module (right)

Fig. 4. Shift-convolution

Fig. 5. Diagrams of P-MHSA. (a) Self-attention mechanism; (b) multi-head attention (MHA); (c) P-MHSA

Fig. 6. Image reconstruction. (a) Overall module; (b) schematic of sub-pixel convolution layer

Fig. 7. Image down-sampling (DSF: down-sampling factor)

Fig. 8. Process and visualization of CutBlur

Fig. 9. Loss function curve

Fig. 10. Super-resolution reconstruction images of a normal retinal OCT image by different models. (a) HR image; (b) TESR reconstructed image; (c) detail of HR image; (d)–(h) local reconstruction effect of SRGAN, RCAN, IPT, SwinIR and TESR

Fig. 11. Super-resolution reconstruction results of a pathological retina OCT images by different models. (a) HR image; (b) TESRreconstructed image; (c) detail of HR image; (d)-(h) local reconstruction effect of SRGAN, RCAN, IPT, SwinIR and TESR