LoHalo Vs Cubic

[sillyxone]

To resize my full-res JPG to share/post on web, I’ve tested out Cubic, Nohalo, Lohalo a while back, and in the end decided to use ImageMagick’s Lanczos resampling after a lot of pixel-peeping. For example, resizing to 800px wide:

/usr/bin/convert input_file.jpg -filter Lanczos -resize 800x -unsharp 1.5x1+0.4+0.02 -quality 90 "output_file.jpg"

I found Lanczos excellent at keeping diagonal lines smooth (instead of jagged), and retaining details and clarity. You can also adjust its strength with the -unsharp param. I used it too frequently that I created a wrapper Python script:

downsampling script (wrapper for Lanczos resampling)

Here’s a Python wrapper script for ImageMagick to resize an image using Lanczos resampling with mild (un)sharpening. This is mostly useful for pixel-peepers Please don’t ask me about the details, I have no idea how Lanczos works, I only know that the result is superior than, say, resizing using standard bicubic in GIMP (retains more details, diagonal lines are less jagged, …). This script mainly is to mass resize a bunch of images at once (and also because ImageMagick’s syntax for dimens…

I also just learned recently that Nohalo is a derivative of Lanczos. At least I did a blind test first

For darktable, I don’t apply sharpening, but use Richardson-Lucy deconvolution at the end as part of my nind-denoise workflow instead.