Fig. 1. The process of ultra-widefield retinal image stitching

Fig. 2. Color fundus image and each channel component. (a) Color fundus image; (b) red channel; (c) green channel; (d) blue channel

Fig. 3. Green channel fundus images to be processed and CLAHE processed results

Fig. 4. Candidate point detection schematic

Fig. 5. Coarse matching results of feature points

Fig. 6. Accurate matching results of feature points

Fig. 7. Two images to be stitched in the panoramic view coordinate system

Fig. 8. Binary mask of the images to be stitched

Fig. 9. Binary mask of overlapping area. (a) Two geometric centers of a binary mask; (b) the overlapping mask after rotation

Fig. 10. Linear weight mask of the overlapping region after rotation. (a) Left side linear weight mask; (b) right side linear weight mask

Fig. 11. Mask generation results. (a) Left mask; (b) right mask

Fig. 12. Stitching of multi-angle images with different exposure differences. (a)‒(c) Stitching of low exposure differential images;

Fig. 13. Stitching of the left and right fundus images from the same physical examination personnel. (a) Four images of the left eye;

Fig. 14. Stitched images of three fusion algorithms. (a)‒(c) Maximum fusion algorithm; (d)‒(f) gradual in and gradual out fusion algorithm; (g)‒(i) our algorithm

Fig. 15. Local entropy images of stitched images of three fusion algorithms. (a)‒(c) Maximum fusion algorithm; (d)‒(f) gradual in and gradual out fusion algorithm; (g)‒(i) our algorithm

Fig. 16. Stitched images of three fusion algorithms and the mask of overlapping region. (a) (b) Maximum fusion algorithm; (c)(d) gradual in and gradual out fusion algorithm; (e)(f) our algorithm

Fig. 17. Original image and reference image. (a) Original image; (b) reference image