Add BM3D(sparse 3D transform-domain collaborative filtering) denoising algorithm.

[Longhao-Chen]

No description provided.

[johnnychen94]

Thanks! Any reason to make it a conditional dependency? (FYI, usually, we use https://github.com/JuliaPackaging/Requires.jl for conditional codes)

Can you also add a demo similar to https://github.com/JuliaImages/ImageNoise.jl/blob/master/docs/examples/reduce_noise/nonlocal.jl?

Also, it would be great to also have a regression test to make sure future improvement don't make it worse, like how

ImageNoise.jl/test/ReduceNoise/NonlocalMean.jl

Lines 62 to 72 in 3381164

	@testset "Numeric" begin
	img_gray = n0f8.(imresize(testimage("lena_gray_256"); ratio=0.25))
	n = AdditiveWhiteGaussianNoise(0.05)
	noisy_img = apply_noise(img_gray, n; rng=MersenneTwister(0))
	f = NonlocalMean(0.05, 2)
	denoised_img = reduce_noise(noisy_img, f)
	# further modification shall not decrease psnr and ssim
	assess(PSNR(), denoised_img, img_gray) >= 28.46
	assess(SSIM(), denoised_img, img_gray) >= 0.918
	end

does.

[Longhao-Chen]

Could you tell me what the role of conditional dependency is？

[johnnychen94]

• hard dependency: directly add BM3DDenoise to Project.toml and let Pkg to manage it
• conditional dependency: don't let Project.toml manage it, and only load it under some circumstance. In the current version, BM3DDenoise will only be loaded when reduce_noise(img, BM3D()) is called.

Generally, we should use hard dependency and let Pkg do everything for us, because it allows 1) less code management, and 2) precompilation.

In some typical cases like FileIO there are too many possible IO backends, we use conditional loading to 1) don't run into a gigantic Project.toml because not everyone needs backend support for every file format, 2) reduce loading time

There're some discussions in JuliaLang/Pkg.jl#977 to allow Pkg do conditional dependency management.

---

In our case, I think we can just use the hard dependency as if BM3DNoise is a normal package.

[Longhao-Chen]

I don't think everyone will use this function. For example, if I just want to use apply_ noise, then I don't have to install BM3DDenoise.jl , which can save disk space. Just like FileIO .
As for precompiling, I did an experiment. Precompiling will be performed during BM3DDenoise.jl installation.

[johnnychen94]

I realize that this package is a bit old and broken, so I've made a PR in #22 to upgrade it. (Just merged)

I don't think everyone will use this function.

Okay, in this case, we can use Requires.jl for this purpose. We can write BM3DDenoise.jl in the usual way, and then conditionally load this file using Requires.

    # in src/ReduceNoise/ReduceNoise.jl
    using Requires

    function __init__()
        @require BM3DDenoise="95fb3b36-088a-43fb-bb1b-b1f34fadbd7d" include("BM3DDenoise.jl")
    end

[johnnychen94]

Oh sorry! Forget about the Requires stuff. I didn't get your ideas correctly. This is a nice idea and I think we should propose to other packages as well.

[johnnychen94]

Let's just add some basic demo and basic tests. And add BM3DDenoise to Project.toml

[Longhao-Chen]

Let's just add some basic demo and basic tests. And add BM3DDenoise to Project.toml

OK, I will.

[codecov]

Codecov Report

Merging #21 (c7f9cae) into master (2cf0300) will increase coverage by 0.61%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master      #21      +/-   ##
==========================================
+ Coverage   93.61%   94.23%   +0.61%     
==========================================
  Files           5        6       +1     
  Lines          94      104      +10     
==========================================
+ Hits           88       98      +10     
  Misses          6        6              

Impacted Files	Coverage Δ
src/ReduceNoise/BM3DDenoise.jl	100.00% <100.00%> (ø)

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 2cf0300...c7f9cae. Read the comment docs.

[johnnychen94]

@Longhao-Chen thank you for adding this!