Demosaicking of Noisy Bayer-Sampled Color Images with Least-Squares Luma-Chroma Demultiplexing and Noise Level Estimation: Additional Results

Eric Dubois and Gwanggil Jeon

IEEE Transactions on Image Processing. Text of accepted manuscript.

Software and images are provided so that results can be reproduced by the reader. Updated August 1, 2012.

Download

Metrics download: CMSE, CPSNR, and S-CIELAB calculators.

Software download: LSLCD_NE.zip.

Demosaicked images with different methods

This web page provides the 4 original and demosaicked images (#1, #8, #19, #24) corresponding to TABLE V and TABLE VI in the paper. Images are provided in TIFF format (uncompressed) and in JPG format (saved with ‘quality’ 90 in MATLAB). Each cell shows the CPSNR (dB) and the S-CIELAB metric when added noise standard deviation parameter is "0, 4, 8, 12, 16 and 20." Methods A to M are described below the table. Selected detail views are shown at the bottom of this page.

No. (thumb)

noise level (sigma)

Original

A

B

C

D

E

F

G

H

I

J

K

L

M

24

1 (0)

TIFF JPG

TIFF JPG 35.48 1.022

TIFF JPG 35.49 1.122

TIFF JPG 33.70 1.601

TIFF JPG 34.51 1.168

TIFF JPG 35.21 1.071

TIFF JPG 35.29 1.055

TIFF JPG 35.01 1.043

TIFF JPG 35.57 1.001

TIFF JPG 35.62 0.973

TIFF JPG 35.68 0.969

TIFF JPG 35.71 0.968

TIFF JPG 35.72 0.967

TIFF JPG 35.68 0.969

3 (4)

TIFF JPG

TIFF JPG 33.65 1.680

TIFF JPG 33.58 1.755

TIFF JPG 32.40 2.125

TIFF JPG 33.12 1.668

TIFF JPG 33.72 1.686

TIFF JPG 33.77 1.679

TIFF JPG 33.34 1.727

TIFF JPG 33.88 1.651

TIFF JPG 33.93 1.634

TIFF JPG 33.96 1.631

TIFF JPG 33.98 1.627

TIFF JPG 34.00 1.626

TIFF JPG 32.88 1.727

5 (8)

TIFF JPG

TIFF JPG 31.28 2.547

TIFF JPG 31.24 2.554

TIFF JPG 30.53 2.794

TIFF JPG 31.01 2.359

TIFF JPG 31.70 2.421

TIFF JPG 31.76 2.401

TIFF JPG 31.07 2.633

TIFF JPG 31.66 2.450

TIFF JPG 31.77 2.396

TIFF JPG 31.81 2.379

TIFF JPG 31.83 2.369

TIFF JPG 31.84 2.365

TIFF JPG 29.94 2.630

7 (12)

TIFF JPG

TIFF JPG 29.37 3.425

TIFF JPG 29.46 3.315

TIFF JPG 29.03 3.379

TIFF JPG 29.22 3.050

TIFF JPG 30.03 3.120

TIFF JPG 30.14 3.045

TIFF JPG 29.23 3.536

TIFF JPG 29.89 3.223

TIFF JPG 30.05 3.104

TIFF JPG 30.13 3.051

TIFF JPG 30.16 3.025

TIFF JPG 30.18 3.012

TIFF JPG 27.94 3.448

9 (16)

TIFF JPG

TIFF JPG 27.87 4.294

TIFF JPG 28.08 4.050

TIFF JPG 27.87 3.901

TIFF JPG 27.82 3.694

TIFF JPG 28.71 3.795

TIFF JPG 28.89 3.636

TIFF JPG 27.81 4.407

TIFF JPG 28.51 3.965

TIFF JPG 28.73 3.773

TIFF JPG 28.85 3.673

TIFF JPG 28.90 3.618

TIFF JPG 28.93 3.590

TIFF JPG 26.56 4.186

11 (20)

TIFF JPG

TIFF JPG 26.64 5.154

TIFF JPG 26.97 4.757

TIFF JPG 26.95 4.372

TIFF JPG 26.68 4.292

TIFF JPG 27.64 4.450

TIFF JPG 27.90 4.192

TIFF JPG 26.69 5.240

TIFF JPG 27.41 4.679

TIFF JPG 27.66 4.412

TIFF JPG 27.84 4.259

TIFF JPG 27.90 4.170

TIFF JPG 27.95 4.120

TIFF JPG 25.56 4.861

Description of the methods used in the above table

Method A: JDD (D. Menon and G. Calvagno, “Joint demosaicking and denoising with space-varying filters,” in IEEE International Conference on Image Processing (ICIP) 2009, Sep. 2009, pp. 477–480.)

Method B: Zhang1 (L. Zhang, X. Wu, and D. Zhang, “Color reproduction from noisy cfa data of single sensor digital cameras,” IEEE Trans. Image Process., vol. 16, no. 9, pp. 2184–2197, Sep. 2007.)

Method C: LASIP (D. Paliy, V. Katkovnik, R. Bilcu, S. Alenius, and K. Egiazarian, “Spatially adaptive color filter array interpolation for noiseless and noisy data,” International Journal of Imaging Systems and Technology, vol. 17, no. 3, pp. 105–122, Oct. 2007.)

Method D: Zhang2 (L. Zhang, R. Lukac, X. Wu, and D. Zhang, “PCA-based spatially adaptive denoising of cfa images for single-sensor digital cameras,” IEEE Trans. Image Process., vol. 18, no. 4, pp. 797–812, Apr. 2009.)

Method E: Condat 9x9 filter with BM3D (L. Condat, “A new color filter array with optimal properties for noiseless and noisy color image acquisition,” IEEE Trans. Image Process., vol. 20, no. 8, pp. 2200–2210, 2011.)

Method F: Condat 13x13 filter with BM3D (L. Condat, “A new color filter array with optimal properties for noiseless and noisy color image acquisition,” IEEE Trans. Image Process., vol. 20, no. 8, pp. 2200–2210, 2011.)

Method G: 5x5 GT filter with BM3D

Method H: 7x7 GT filter with BM3D

Method I: 9x9 GT filter with BM3D

Method J: 11x11 GT filter with BM3D

Method K: 13x13 GT filter with BM3D

Method L: 15x15 GT filter with BM3D

Method M: 11x11 GT filter with h_{LN}

Section VI-C additional results

CPSNR (M₁) and S-CIELAB (M₂) metrics for demosaicked images using different noise estimation schemes.

Figures

(a) Comparison of noise estimation methods in the sense of CPSNR (dB).

(b) Comparison of noise estimation methods in the sense of CPSNR (dB): difference from BF(σ_A).

(c) Comparison of noise estimation methods in the sense of S-CIELAB.

(d) Comparison of noise estimation methods in the sense of S-CIELAB: difference from BF(σ_A).

Table

Section VI-D additional results

We present here zoomed detail views (with bicubic interpolation) of selected areas in the four images shown above and two additional images, with the three leading demosaicking methods, and two levels of noise: ANL = 5 (sigma_A = 8) and ANL = 11 (sigma_A=20).

Detail of #1 with ANL = 5; sigma_A = 8

Original	noisy CFA image with sigma_A = 8
Zhang2009	Condat2011	LSLCD_NE

Detail of #1 with ANL = 11; sigma_A = 20

Original	noisy CFA image with sigma_A = 20
Zhang2009	Condat2011	LSLCD_NE

Detail of #8 with ANL = 5; sigma_A = 8

Original	noisy CFA image with sigma_A = 8
Zhang2009	Condat2011	LSLCD_NE

Detail of #8 with ANL = 11; sigma_A = 20

Original	noisy CFA image with sigma_A = 20
Zhang2009	Condat2011	LSLCD_NE

Detail of #19 with ANL = 5; sigma_A = 8

Original	noisy CFA image with sigma_A = 8
Zhang2009	Condat2011	LSLCD_NE

Detail of #19 with ANL = 11; sigma_A = 20

Original	noisy CFA image with sigma_A = 20
Zhang2009	Condat2011	LSLCD_NE

Detail of #24 with ANL = 5; sigma_A = 8

Original	noisy CFA image with sigma_A = 8
Zhang2009	Condat2011	LSLCD_NE

Detail of #24 with ANL = 11; sigma_A = 20

Original	noisy CFA image with sigma_A = 20
Zhang2009	Condat2011	LSLCD_NE

Detail of Bike image with ANL = 5; sigma_A = 8

Original	noisy CFA image with sigma_A = 8
Zhang2009	Condat2011	LSLCD_NE

Detail of bike image with ANL = 11; sigma_A = 20

Original	noisy CFA image with sigma_A = 20
Zhang2009	Condat2011	LSLCD_NE

Detail of Spincalendar image with ANL = 5; sigma_A = 8

Original	noisy CFA image with sigma=8
Zhang2009	Condat2011	LSLCD_NE

Detail of Spincalendar image with ANL = 11; sigma_A = 20

Original	noisy CFA image with sigma_A = 20
Zhang2009	Condat2011	LSLCD_NE