Fig 1.
Contourlet Filter bank consisting of Laplacian Pyramid (LP) and Directional Filter Bank (DFB) [39].
Fig 2.
The comparison of sparsity between wavelet and contourlet transforms on the Lena image.
(a) wavelet, (b) contourlet.
Fig 3.
Parent-child relationship of contourlet coefficients.
(a) a parent coefficient and its pertinent four child coefficients (b) parent-child dependency of contourlet coefficients in log scale.
Fig 4.
Block diagram for the proposed method (CT: contourlet transform).
Fig 5.
Plots of noise variances in the wavelet domain for the 50th row of the Lena image; noise variances by Eq (12) (blue solid line), sample noise variances from 100 times of noise generation (green dashed line), 1,000 times of noise generation (red dotted line) and noise variance estimated in the image domain (cyan dash-dot line).
Table 1.
Denoising performance evaluation of each block shown in Fig 4 in terms of PSNR (dB) for various noise ratio of Poisson and Gaussian components with fixed noise variance.
Fig 6.
Denoising results on Lena image (http://en.wikipedia.org/wiki/File:Lenna.png.) with a = 1/3 and b = 0.32.
(a) original image, (b) noisy image (7.48 dB), (c) BM3D (24.80 dB), (d) PURE-LET (24.68 dB), (e) the proposed method (24.72 dB).
Fig 7.
Denoising results on the magnified Lena image with a = 1/3 and b = 0.32.
(a) original image, (b) noisy image, (c) BM3D, (d) PURE-LET, (e) the proposed method.
Table 2.
Comparison of the proposed method with the BM3D method and PURE-LET in terms of PSNR (dB) for various low-count cases.
Fig 8.
Denoising results on the first set of HeLa cell images.
(a) an average of 100 images used as the ground truth, (b) single acquisition image, (c) BM3D, (d) the proposed method.
Table 3.
Comparison of the proposed method with the BM3D method in terms of PSNR (dB) for each channel on the first set of HeLa cell images.
Fig 9.
The second set of HeLa cell images and denoising results.
First low: the second set of HeLa cell images with different laser intensities. (a) 0.2, (b) 0.4, (c) 0.8. Second low: denoising results on HeLa cell image with laser intensity 0.4. (d) an average of 40 images used as the ground truth, (e) BM3D, (f) the proposed method.
Table 4.
Comparison of the proposed method with the BM3D method in terms of PSNR(dB) for various laser intensities on the second set of HeLa cell images.