Table 1 Comparison of denoising performance between BM3D filter and deep learning models.

From: Improvement of nerve imaging speed with coherent anti-Stokes Raman scattering rigid endoscope using deep-learning noise reduction

Model

Evaluation metrics

PSNR

SSIM

BM3D

\(31.04 \pm 5.86\)

\(0.649 \pm 0.207\)

DN

\(32.05 \pm 6.02\)

\(0.671 \pm 0.220\)

N2N

\(\mathbf {32.21 \pm 6.04}\)

\(\mathbf {0.679 \pm 0.214}\)

W5

\(31.96 \pm 6.02\)

\(0.668 \pm 0.221\)

Paired t-test

P-value

PSNR

SSIM

N2N vs BM3D

0.045*

0.016*

N2N vs DN

0.0059*

0.027*

N2N vs W5

0.00035*

0.016*

  1. The average and standard deviation of each metric over five test images are shown. The N2N shows the significant difference from the others in a paired t-test (\(n=5\), \(\mathrm {P} < 0.05\)) using the Bonferroni–Holm correction. N2N showed the highest performance for both metrics. * means significant difference after the Bonferroni–Holm correction (\(\alpha < 0.05\)).
Back to article page