[EXPERIMENTAL] v.colors: Histogram equalization

[HuidaeCho]

Lightness (in HSL)-based color histogram equalization. Calculates probability using feature areas (no cells in vector), but I may not merge it if I fail to fix the following issue.

v.colors map=censusblk_swwake use=attr column=TOTAL_POP color=blues -e

seems to work,

but

v.colors map=censusblk_swwake use=attr column=TOTAL_POP color=rainbow -e

doesn't work.

[HuidaeCho]

Method

1. RGB => HSL
2. Sort areas by L in an ascending order
3. For each area, calculate the cumulative area (sum_area) and probability (L) by sum_area / total_area
4. Original (H, S) and new L from step 3 => RGB

Issues This method seems to work for monotonic color maps, but a color map such as rainbow has only two lightnesses 0.5 and 1 and it doesn't really make much sense to histogram equalize two intensities. Maybe, I should do color-averaged equalization.

[HuidaeCho]

In fact, these two commands don't work either:

v.colors map=censusblk_swwake use=attr column=TOTAL_POP color=blues -g
v.colors map=censusblk_swwake use=attr column=TOTAL_POP color=blues -a

because Rast_log_colors() and Rast_abs_log_colors() interpolates categories, which doesn't make sense for vector maps. Vector maps require exact categories in the color file, it seems?

[cmbarton]

This does not work correctly on my Mac. I just cloned and complied it within the past hour. I tested it with the NC census block data. A good place to look is with the TOTAL_POP attribute because it has an area much higher than any others: cat=1319. Checking the histogram equalization box gives unexpected and incorrect results with a multi-color table (check out one of the viridis tables or GRASS bgyr). It seems initially to work with a single color gradient like blues or reds. But even these are incorrect. TOTAL_POP for cat=1319 should be the most intense color but it is not. It looks exactly the same as some other areas like cat=1364. You can try the same equalizations on a raster version of the census block vector area map and it looks correct.

[HuidaeCho]

This does not work correctly on my Mac. I just cloned and complied it within the past hour. I tested it with the NC census block data. A good place to look is with the TOTAL_POP attribute because it has an area much higher than any others: cat=1319. Checking the histogram equalization box gives unexpected and incorrect results with a multi-color table (check out one of the viridis tables or GRASS bgyr). It seems initially to work with a single color gradient like blues or reds. But even these are incorrect. TOTAL_POP for cat=1319 should be the most intense color but it is not. It looks exactly the same as some other areas like cat=1364. You can try the same equalizations on a raster version of the census block vector area map and it looks correct.

I see. I think this PR needs some (a lot of?) work. It gets tricky to implement histogram equalization for vector features. BTW, have you tried -a or -g? Do they work for you?

v.colors map=censusblk_swwake use=attr column=TOTAL_POP color=blues -a
v.colors map=censusblk_swwake use=attr column=TOTAL_POP color=blues -g

[HuidaeCho]

I see. I think this PR needs some (a lot of?) work. It gets tricky to implement histogram equalization for vector features. BTW, have you tried -a or -g? Do they work for you?

v.colors map=censusblk_swwake use=attr column=TOTAL_POP color=blues -a
v.colors map=censusblk_swwake use=attr column=TOTAL_POP color=blues -g

Nope. They don't work (https://github.com/OSGeo/grass/issues/1432#issuecomment-804281748).

[neteler]

@HuidaeCho would you mind to rebase this PR?

[HuidaeCho]

@HuidaeCho would you mind to rebase this PR?

@neteler Please try it.

[neteler]

I have tested it again successfully with the following test:

# main
grass ~/grassdata/nc_spm_08_grass7/user1
g.region vector=censusblk_swwake -p
d.mon  wx0

# show original, suboptimal color table
d.vect censusblk_swwake

# apply histogram equalization
v.colors map=censusblk_swwake use=attr column=TOTAL_POP color=blues -e

# show with improved color table
d.vect censusblk_swwake 

# restore original color table
v.colors map=censusblk_swwake use=attr column=TOTAL_POP color=blues 

d.redraw

Which edge cases shall be tested?