Imagemagick maximum colors and scaling - resize

Question

I am trying to convert 80x80 images to 56x56 images in greyscale in 2bpp.
The 80x80 images are colored and probably have up to 16 colors in them. They also have a transparent background.
I need them greyscaled with 4 colors in them, with white being the lightest and black being the darkest.
Each image has multiple colors in it, but each color has a palette of 3 colors, a dark one, medium one, and light one.
I need to convert all the dark ones to a dark grey, the medium ones to a light grey, and the light ones to white, while maintaining the black that is already in the image.
I can successfully convert the image to greyscale, trim the canvas, and fill the background with this command
convert input.png +dither -flatten -trim -set colorspace Gray -
separate -average output.png
Now I need to limit the colors, but it's not converting the right ones. The light colors are getting converted to light grey instead of white. When I change -level option, it only works with some of the images.
-auto-levels does not do what I want either.
Is there anyway to set the colors in the midrange to automatically level themselves to fit my requirements? I'm sorry if I'm not explaining this enough.
This is the code I've been tampering with but it only works on a few images. Messing with the gamma option can make it work for more images but breaks the original working images.
convert "$f" +dither -flatten -trim -set colorspace Gray -separate -
average -gamma 1.37 -level 25%,75% -colors 4 -adaptive-resize 56x56\>
-background white -gravity center -extent 56x56 -remap nido.png
"${f%.png}".png2
I can't provide an expected image, but an image something similar to what is expected. Here is original image https://img.pokemondb.net/sprites/black-white/normal/charizard.png and here is desired output format image https://img.pokemondb.net/sprites/red-blue/normal/charizard.png
Here is what I've got so far
https://www.pokecommunity.com/showthread.php?p=9692599#post9692599
convert "$f" +dither -background white -flatten -trim -adaptive-resize
56x56\> "${f%.png}".png2
convert "${f%.png}".png2 +dither -colorspace gray -separate -average
"${f%.png}".png2
convert "${f%.png}".png2 +dither -gamma 3.0 -black-threshold 70%
"${f%.png}".png2
convert "${f%.png}".png2 +dither -gamma 0.45 -white-threshold 90%
"${f%.png}".png2
convert "${f%.png}".png2 +dither -remap nidoking.png -background white
-gravity center -extent 56x56 "${f%.png}".png2
Btw, ^ is in a for loop, hence the variables. Changing the gamme and black/white theshold values are getting me closer, but it is extremely tedious and when I get one image to correctly convert another breaks. nidoking.png is my remap file. remap works perfect, it's just before the remap the colors are being separated or filtered properly.
Solved, thanks to Mark Setchell
This is what I ended up doing
#!/bin/bash
rm x*
rm colors*
cd images
rm *.png2
rm *.txt
for f in *.png
do
#Fitting canvas to image, setting background color, and removing transparency
convert "$f" +dither -background white -flatten -trim "${f%.png}".png2
#Converting image to greyscale
convert "${f%.png}".png2 +dither -colorspace gray -separate -average "${f%.png}".png2
#Resizing without blurring/adding pixels
convert "${f%.png}.png2" +dither -interpolate Nearest -interpolative-resize 56x56\> "${f%.png}".png2
#Grabbing every color used in image and putting it in a text file
convert "${f%.png}.png2" txt: | sed '1d' | cut -f 4 -d " " | sort -u > "${f%.png}".txt
done
#Putting all colors into one file
cat *.txt >> ../colors
cd ../
#One last clean up of file/sorting
cat colors | tr " " "\n" | sort -u > colors.txt
rm colors
#Splitting the hex codes into four files for each desired color
file=colors.txt
lines=$(wc -l <${file})
((lpp = (lines + 4 - 1) / 4))
split --lines=${lpp} ${file}
#Going back to images directory
cd images
for f in *.png
do
#Each while loop reads everyone line of the specified file and puts it in variable $i, then I use $i to convert to one of the desired 4 colors.
cat ../xaa | while read i
do
convert "${f%.png}".png2 +dither -fuzz 0% -fill "#000000" -opaque "${i}" "${f%.png}".png2
done
cat ../xab | while read i
do
convert "${f%.png}".png2 +dither -fuzz 0% -fill "#555555" -opaque "${i}" "${f%.png}".png2
done
cat ../xac | while read i
do
convert "${f%.png}".png2 +dither -fuzz 0% -fill "#AAAAAA" -opaque "${i}" "${f%.png}".png2
done
cat ../xad | while read i
do
convert "${f%.png}".png2 +dither -fuzz 0% -fill "#FFFFFF" -opaque "${i}" "${f%.png}".png2
done
mv "${f%.png}".png2 ../finished/"${f}"
done
This script turned this
Into this

Answer 1

Basically, the idea would be to do a reduction to 4 colours in the RGB colourspace (rather than in greyscale colourspace) to get the best four colours. Then get the lightness of each of those and map the darkest one to black, the next lighter to dark grey, the next lighter to light grey and the lightest to white.
Here it is mapped to the 4 best colours in RGB colourspace:
The code, without much error checking or corner-case handling, looks like this:
#!/bin/bash -x
# Do a colour reduction to get best 4 colours in RGB colourspace rather than in grey colourspace
magick pokething.png -alpha off +dither -colors 5 -unique-colors unique.png
# Get hex colours into array "hexcolours[]"
hexcolours=( $(convert unique.png txt: | awk 'NR>1{print $3}') )
echo DEBUG: hexcolours=${hexcolours[#]}
# Get lightness of each colour into array "lightness[]", i.e. H of HSL
# Note ggrep is just GNU grep
lightness=( $(convert unique.png -colorspace HSL -separate -delete 0,1 txt: | ggrep -Po "\d+(?=\)$)") )
echo DEBUG: lightness=${lightness[#]}
# Sort the colours by their lightness
fourshades=( $(for ((i=0;i<4;i++)) ;do
echo ${lightness[i]} ${hexcolours[i]}
done | sort -n | awk '{print $2}') )
echo DEBUG: fourshades=${fourshades[#]}
# Now change those colours in original image
magick pokething.png -alpha off +dither -colors 5 -fuzz 10% \
-fill black -opaque "${fourshades[0]}" \
-fill gray25 -opaque "${fourshades[1]}" \
-fill gray75 -opaque "${fourshades[2]}" \
-fill white -opaque "${fourshades[3]}" \
result.png
The output is as follows:
DEBUG: hexcolours=#000000 #094152 #A95244 #EF9E3C
DEBUG: lightness=0 46 119 150
DEBUG: fourshades=#000000 #094152 #A95244 #EF9E3C
That results in this being executed:
magick pokething.png -alpha off +dither -colors 5 -fuzz 10% \
-fill black -opaque '#000000' \
-fill gray25 -opaque '#094152' \
-fill gray75 -opaque '#A95244' \
-fill white -opaque '#EF9E3C' result.png
So, basically I am replacing #094152 with dark grey because 46 is the second darkest colour present. Then I am replacing #A95244 with light grey because 119 is the next lighter colour, then replacing #EF9E3C with white because that is the lightest colour.

Answer 2

You might be able to get your desired result by using pretty much only "-remap". This example will make a four color map from white, 65% gray, 35% gray, and black, and write that map to temporary memory. Next it reads your input image, sets the background to white, and flattens the image. Then it turns off dithering, remaps the input image to the map you created, and returns the result.
convert xc:white xc:gray65 xc:gray35 xc:black -append -write mpr:map +delete \
input.png -background white -flatten +dither -remap mpr:map output.png
You should do any resizing, cropping, etc., before the remap, of course. You probably don't need to turn it to grayscale first because the map has no color.