Legacy ImageMagick Discussions Archive

New IM user and first post.
IM 7.0.7-19 Q16 x64 on a Windows 10 64-bit machine

I'm building a Windows batch script to handle downsizing of high resolution B&W TIFF files (originally produced on my drum scanner). The source files are TIFF, 16-bit, in Gray Gamma 2.2 colorspace. Output files will have the same attributes: TIFF, 16-bit, in Gray Gamma 2.2 colorspace (no change from input).

Here's an example of how I am currently doing the resize:

magick -quiet "filename.tif" -filter Quadratic -distort Resize 10800 -set density 720x720 "filename_720.tif"

This downsizes to a new width of 10800 pixels (height calculated by IM) and sets a new resolution of 720ppi into the output image. This command works fine, producing a new TIFF file with the desired 16-bit depth and Gray Gamma 2.2 colorspace ( as verified with "magick identify").

In Photoshop, a comparison of the source and output files shows image tonal values appear to have been extremely well preserved (though I will need to add some small amount of sharpening). However, I'm wondering whether I should be converting to a linear colorspace before doing the resize, then convert back. For color images (multi-channel) it's clear this is recommended. But for grayscale TIFF files, is it also recommended to do the colorspace conversion? What am I giving up if I don't do the conversion?

Assuming I should do the conversion, which of these would be better:

magick -quiet "filename.tif" -filter Quadratic -gamma 0.454545 -distort Resize 10800 -gamma 2.2 . . .
or
magick -quiet "filename.tif" -filter Quadratic -colorspace LinearGray -distort Resize 10800 -colorspace Gray . . .

Thanks,
Dave

To be technically correct, you should resize in linear colorspace. Does it make any difference? I doubt it. The difference will show only where you have very high frequency data (ie rapidly changing pixels).

I haven't tried the new "-colorspace LinearGray" yet, but that's probably the correct way. Doing it with gamma will be only very slightly wrong.

You mentioned the need to sharpen after downsizing. That should have triggered me to mention the work by Nicolas Robidoux and others, which does both operations together, by blending a linear resize with one with a further shift of gamma 3. The results are excellent.

See Resampling with halo minimization, and the links from there.