Denoising Algorithm of Multi-Pinhole Collimated X-Ray Fluorescence CT Based on Noise Level Estimation

Ruge Zhao; Peng Feng; Yan Luo; Song Zhang; Peng He; Yanan Liu

doi:10.3788/AOS230679

Acta Optica Sinica, Volume. 43, Issue 20, 2034001(2023)

Denoising Algorithm of Multi-Pinhole Collimated X-Ray Fluorescence CT Based on Noise Level Estimation

Ruge Zhao¹, Peng Feng^1,2、*, Yan Luo¹, Song Zhang¹, Peng He^1,2, and Yanan Liu^3、**

Author Affiliations

¹Key Lab of Optoelectronic Technology & Systems, Ministry of Education, Chongqing University, Chongqing 400044, China

²ICT NDT Engineering Research Center, Ministry of Education, Chongqing University, Chongqing 400044, China

³School of Electronics and Information Engineering, Chongqing Technology and Business Institute, Chongqing 400032, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(14)

Fig. 1. Schematic diagram of XFCT imaging

Download full size

Fig. 2. Flow chart of NeCNN denoising algorithm

Download full size

Fig. 3. Setup of XFCT imaging system

Download full size

$Two kinds of phantoms. (a) Physical object of PMMA phantom; (b) phantom 1: phantom with high mass fraction; (c) phantom 2: phantom with low mass fraction$

Fig. 4. Two kinds of phantoms. (a) Physical object of PMMA phantom; (b) phantom 1: phantom with high mass fraction; (c) phantom 2: phantom with low mass fraction

Download full size

Fig. 5. XFCT images from simulation. (a) Noise image; (b) standard image (clean image)

Download full size

Fig. 6. Denoising results under different noise levels. (a), (e), (i) are original images (testing images), and σ_a=14.0525, σ_e=20.7962, and σ_i=31.7817; (b), (f), (j) are denoising images of NeCNN; (c), (g), (k) are denoising images of DnCNN; (d), (h), (l) are denoising images of BM3D

Download full size

Fig. 7. Denoising result of PMMA phantom. (a) Testing image; (b) denoising image of NeCNN

Download full size

Fig. 8. Profile of pixel value distribution with different denoising algorithms. (a) Noise image; (b) denoising image of BM3D; (c) denoising image of NeCNN

Download full size

Table 1. NeCNN hyperparameters
View table
Table 1. NeCNN hyperparameters
Hyperparameter Value
Kernel size 3×3
Resolution 420×560
Num epoch 30
Layer 17
Learning rate 0.001
Optimizer Adam
Batch size 8
Activation function ReLU

Table 2. Dataset parameter settings
View table
Table 2. Dataset parameter settings
Phantom Air phantom PMMA phantom
Image number 400 100
Format .jpg .jpg
Resolution 420×560 420×560
Number of training set 320 80
Number of test set 80 20

Table 3. DnCNN parameters
View table
Table 3. DnCNN parameters
Parameter Value
Kernel size 3×3
Resolution 420×560
Num epoch 30
Layer 17
Learning rate 0.001
Optimizer Adam
Batch size 8
Activation function ReLU

Table 4. BM3D parameters
View table
Table 4. BM3D parameters
Parameter Value
Size of block 20×20
Number of max matching block 16
Threshold of block-wise similarity（step 1） 250000
Threshold of block-wise similarity（step 2） 3600

Table 5. Comparison of denoising results with different algorithms
View table
Table 5. Comparison of denoising results with different algorithms
PSNR /dB SSIM
Original image 27.75589 0.76805
NeCNN 29.01558 0.95066
DnCNN 28.77565 0.92332
BM3D 28.39545 0.80807

Table 6. Denoising results of PMMA phantom
View table
Table 6. Denoising results of PMMA phantom
PSNR /dB SSIM
Original Image 30.65151 0.93161
NeCNN 32.00803 0.95843

Tools

Get Citation

Copy Citation Text

Ruge Zhao, Peng Feng, Yan Luo, Song Zhang, Peng He, Yanan Liu. Denoising Algorithm of Multi-Pinhole Collimated X-Ray Fluorescence CT Based on Noise Level Estimation[J]. Acta Optica Sinica, 2023, 43(20): 2034001

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites