This README documents the necessary steps to run our tests.

• A NEW denoising method that combines 3D Non-Local Means and LBP-TOP (local binary patterns on three orthogonal planes)
• Code version: 0.01-alpha
• Learn Markdown

• Python 2.7
• OpenCV (for Python)
• NumPy
• SciPy
• scikit-image
• bob.ip.base (from Bob library). See also this link
• Joblib

The file runNonLocalMeans run a filtering process with the user parameters:

• -in or --input: Input PATH with the input images (With Noise)
• -out or --output: Input PATH to save the output image (Denoised)
• -ori or --originals: Input PATH with the original images (Without Noise)
• -H or --filterStrength: Input the parameter H (Filter Strength)
• -p or --patch: Input the size of patch (Neighborhood size of centered voxel)
• -w or --window:Input the size of search window (Window size of centered voxel)
• -msb or --MSBValue: Value of Most Significant Bit
• -seq or --sequence: Sequence name of test (E.g. seq1)
• -f or --folder: Input PATH with database to save results (.db)

The file runTests.sh run the filtering process with more than one parameter: combining search window size s = {7, 9, 11 and 13} with patch window size p = {7}; search window size s = {5} with patch window size p = {3}; and all previous combinations with filter strength H = {10, 15, 20 and 25}. This file have four parameters:

• -f or --folder: base directory"
• -s or --sigma: sigma value for the gaussian noise"
• -i or --image: frame sequence folder (relative path)"
• -m or --msb: number of MSB to be considered"

Example how to run:

./runTests.sh -f /home/user/resultexp/ -i seq1 -s 10 -m 3

To execute the same examples that we run, execute the script runAllTests.sh. This script runs for each sequence dataset (with different levels of noise) the filtering process.

All the results will be save on table results on the database 3d_denoising_results.db. The best values of filtering process can be get by the following code:

select seq, sigma, method, h, p, w, msb, psnr, max(ssim) 
from results 
where sigma == 25 and seq == 'seq4' and
method in ('NLM2D','NLM2D-LBP','NLM3D','NLM3D_viORI_texLBP_Adaptive','NLM3D_viORI_texLBP-MSB')
group by method 
order by max(ssim) desc;

The tests with BM3D-SAPCA [1] were made through the source code available in Alessandro Foi personal webpage in MATLAB using the same sequences of images. The sigma parameter used was the value of the noise level.

The algorithm can be run with the sample following code:

originalImage = imread('./dataset/seq1/original/gn-0-009625.png');
noiseImage = imread('./dataset/seq1/gaussian_noise-25/gn-25-009625.png');

[na denoisedImage] = BM3D(1, noiseImage, 25);

psnrV = PSNR(denoisedImage, originalImage);
ssimV = ssim(denoisedImage, originalImage);

[1] K. Dabov, A. Foi, V. Katkovnik, and K. Egiazarian, “Bm3d image denoising with shape-adaptive principal component analysis.” in SPARS’09-Signal Processing with Adaptive Sparse Structured Representations, Saint Malo, France, 2009.