I have made custom median filter for color images, but it performs slow and takes high memory compared to OpenCV.medianBlur. Here is code
using ImageIO
using FileIO
using ImageFiltering
using ImageCore
# using Statistics
using OpenCV
using BenchmarkTools
function median_filter(img, ksize::Tuple{Int64,Int64}=(3, 3))
# mapwindow(median!, img, ksize)
mapwindow(ImageFiltering.MapWindow.median_fast!, img, ksize)
end
function color_median_filter(img, ksize::Tuple{Int64,Int64}=(3, 3))
cview_img = channelview(img)
s::Tuple{Int64,Int64,Int64} = size(cview_img)
filtered_image::Array{N0f8,3} = zeros(N0f8, s)
for idx in 1:s[1]
filtered_image[idx, :, :] = median_filter(cview_img[idx, :, :], ksize)
end
colorview(RGB, filtered_image)
end
noisy_img_path = "lighthouse_noisy.png"
lighthouse_noisy_cv = OpenCV.imread(noisy_img_path)
lighthouse_noisy = load(noisy_img_path)
s1 = 3
s2 = (3, 3)
julia> @benchmark OpenCV.medianBlur($(lighthouse_noisy_cv), $(s1))
BenchmarkTools.Trial: 10000 samples with 1 evaluation.
Range (min … max): 423.100 μs … 1.064 ms ┊ GC (min … max): 0.00% … 0.00%
Time (median): 460.600 μs ┊ GC (median): 0.00%
Time (mean ± σ): 476.162 μs ± 48.744 μs ┊ GC (mean ± σ): 0.00% ± 0.00%
█▅▂▄▇▃
▁▂▃▃▆███████▅▄▄▃▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▁▂▁▁▂▁▁▁▂▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁ ▂
423 μs Histogram: frequency by time 663 μs <
Memory estimate: 1.05 KiB, allocs estimate: 37.
julia> @benchmark color_median_filter($lighthouse_noisy, $s2)
BenchmarkTools.Trial: 41 samples with 1 evaluation.
Range (min … max): 118.774 ms … 134.163 ms ┊ GC (min … max): 0.00% … 3.73%
Time (median): 121.594 ms ┊ GC (median): 0.00%
Time (mean ± σ): 122.864 ms ± 3.296 ms ┊ GC (mean ± σ): 0.94% ± 1.53%
▁ ▁█
▄▄▄▄█▄▄▁██▇▇▁▁▇▇▄▄▁▄▄▁▄▁▁▇▇▁▄▁▁▁▁▇▁▁▁▁▁▄▁▁▁▁▁▁▁▁▄▁▁▁▁▁▁▁▁▁▁▁▄ ▁
119 ms Histogram: frequency by time 134 ms <
Memory estimate: 12.37 MiB, allocs estimate: 107429.
Both filters remove noise from original image. Noisy image created using below code
using Noise
using TestImages
using Images
test_img = testimage("lighthouse")
save("lighthouse_noisy.png", salt_pepper(test_img))
Julia version information
Julia Version 1.9.3
Commit bed2cd540a (2023-08-24 14:43 UTC)
Build Info:
Official https://julialang.org/ release
Platform Info:
OS: Windows (x86_64-w64-mingw32)
CPU: 16 × AMD Ryzen 7 PRO 5850U with Radeon Graphics
WORD_SIZE: 64
LIBM: libopenlibm
LLVM: libLLVM-14.0.6 (ORCJIT, znver3)
Threads: 1 on 16 virtual cores
Thanks
