Fig. 1. Workflow of the dual-cross-correlation-based translation and rotation registration (DCCTRR) algorithm.

Fig. 2. Schematic of our home-built SDOCT system. L1-L5, achromatic lenses; PC1-PC2, polarization controllers; DP, dispersion compensation pair.

Fig. 3. Results of the phantom experiment (silk fabric with a silk fiber diameter of ∼ 32 µm is used to mimic biological microvasculature) with the normalized cross-correlation method. (a) The normalized cross-correlation coefficient map between an image pair. The horizontal coordinate represents the position of the registered image relative to the reference image, and the vertical coordinate represents the cross-correlation coefficient value at that corresponding position. (b) The plot of the image overlap rates versus the normalized SNR of the correlation coefficient map. (c) The plot of different image overlap rates versus normalized registration accuracy. (d)–(h) Merged images of five pairs of representative images with different overlap rates. (i)–(m) Merged RGB images. The red and green channels are the two original binary images, and the matched signaled regions are yellow.

Fig. 4. Results of image registration of cross-correlation method for the images with rotations. (a)–(e) The images with different rotation angles of 0°–4° respectively relative to the same referent image, and the overlap rates are all 96.6%. (f)–(j) The merged RGB images of (a)–(e), respectively. The red and green channels refer to the two original binary images, and the yellow indicates the matched signaled regions. (k) The plot of different rotation angles versus the obtained normalized SNR of the correlation coefficient. (l) The plot of different overlap rates versus the normalized registration accuracy.

Fig. 5. Results of the comparison between the DCCTRR and FMT methods. In the coefficient map, the horizontal coordinate represents the position of the registered image relative to the reference image, and the vertical coordinate represents the correlation coefficient value at that corresponding position. (a) The normalized cross-correlation coefficient map. (b) The phase-correlation coefficient map. (c) The plot of the normalized SNRs of the correlation coefficient map versus different overlap rates. (d) The plot of the normalized registration accuracies versus different overlap rates. (e)–(i) Merged images of five representative overlap rates with the DCCTRR method (the first and the last are the images with the minimum and the maximum overlap rates, respectively). (j)–(n) The merged RGB images of (e)–(i), respectively. (o)–(s) Merged images of the same five pairs of images as (e)–(i) with the FMT method. (t)–(x) The merged RGB images of (o)–(s), respectively.

Fig. 6. DCCTRR and FMT methods for registering images with added Gaussian noise. (a) Registered images without added Gaussian noise. Merged images using the (b) DCCTRR and (c) FMT methods with the maximum STD of the added Gaussian noise. The merged RGB images of (a), (b), and (c) are shown in (d), (e), and (f), respectively. (g) The plot of the normalized SNRs of the coefficient map versus the STDs of the added Gaussian noise. (h) The plot of the normalized registration accuracies versus the STDs of the added Gaussian noise.

Fig. 7. Registration results of human finger OCTA images with the DCCTRR method. (a)–(f) Merged images of six representative overlap rates (The first one and the last one, respectively, have the maximum and the minimum overlap rates). (g)–(l) The corresponding RGB images of (a)–(f), respectively. (m) The plot of the normalized SNRs of the correlation coefficient map versus the different overlap rates. (n) The plot of the normalized registration accuracies versus the different overlap rates.

Fig. 8. Merged large-scale OCTA images with 3 × 3 OCTA en face images. (a) The left hand of a healthy volunteer (the marked region by a solid line was scanned). (b) The merged large-scale image obtained by the DCCTRR method.