using ImageBase.fdiff and some other pre-allocation to improve the performance

[JKay0327]

Some improvement

• ImageBase.fdiff is used to improve the performance, solves https://github.com/johnnychen94/ImageSmooth.jl/issues/4#issue-963142327
• Excluding some unnecessary allocation from codes like similar(float64.(input)) or sum(), solves https://github.com/johnnychen94/ImageSmooth.jl/issues/3#issue-963142254

Previous performance is as follow:

julia> img_gray = testimage("cameraman");

julia> img_rgb = testimage("lena_color_512");

julia> fₛ = L0Smooth()
L0Smooth(0.02, 2.0, 100000.0)

julia> @btime smooth(img_gray, fₛ)
  418.836 ms (1712 allocations: 316.75 MiB)
julia> @btime smooth(img_rgb, fₛ)
  1.155 s (1934 allocations: 1.03 GiB)

Current performance is as follow:

julia> img_gray = testimage("cameraman");

julia> img_rgb = testimage("lena_color_512");

julia> f = L0Smooth()
L0Smooth(0.02, 2.0, 100000.0)

julia> @btime smooth(img_gray, f)
  393.819 ms (1449 allocations: 138.77 MiB)

julia> @btime smooth(img_rgb, f)
  1.013 s (1627 allocations: 387.81 MiB)

We can see that the allocations are extremely reduced, however, the improvement of running time is not as remarkable as the allocations.

Maybe we can do some further improvement, or the main time-consuming factor is the process of fft! and ifft! so that the performance of this algorithm stay at this level.

[codecov[bot]]

Codecov Report

Merging #6 (7741140) into master (500f97c) will increase coverage by 22.52%. The diff coverage is 100.00%.

@@             Coverage Diff              @@
##           master        #6       +/-   ##
============================================
+ Coverage   77.47%   100.00%   +22.52%     
============================================
  Files           4         2        -2     
  Lines         111        69       -42     
============================================
- Hits           86        69       -17     
+ Misses         25         0       -25     

Impacted Files	Coverage Δ
src/SmoothAPI/smooth.jl	100.00% <100.00%> (ø)
src/algorithms/l0_smooth.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 500f97c...7741140. Read the comment docs.

[johnnychen94]

Use squash and merge when you're ready.