Legacy ImageMagick Discussions Archive

IM 6.7.5.6 Q16 Mac OSX Snow Leopard

I believe that -color-matrix should reproduce the following -fx expression. They produce the same results when using values between 0 and 1, but different results when I have values greater than 1 and negative values. I think that the fault is in -color-matrix.

Input:



This works fine -- the results are the same:

rf=0.3
gf=0.6
bf=0.1
convert zelda3.png -separate \
-fx "$rf*u[0]+$gf*u[1]+$bf*u[2]" \
zelda3_gray_mix1a.png




rf=0.3
gf=0.6
bf=0.1
convert zelda3.png -color-matrix "\
$rf 0 0 \
0 $gf 0 \
0 0 $bf \
" -separate -evaluate-sequence add \
zelda3_gray_mix1b.png




But these produce different results and I believe that -color-matrix is the wrong one.

rf=-0.40
gf=1.44
bf=-0.04
convert zelda3.png -separate \
-fx "$rf*u[0]+$gf*u[1]+$bf*u[2]" \
zelda3_gray_mix2a.png




rf=-0.40
gf=1.44
bf=-0.04
convert zelda3.png -color-matrix "\
$rf 0 0 \
0 $gf 0 \
0 0 $bf \
" -separate -evaluate-sequence add \
zelda3_gray_mix2b.png




P.S. I noticed that there was a similar old bug report that might be related. See viewtopic.php?f=3&t=16787

Compile with HDRI and you will get the same result for both commands. Without HDRI you get a difference due to clamping the result between 0 and QuantumRange. The -fx option adds R+G+B then clamps. With -color-matrix, it clamps R, G, and B and only then does it sum R+G+B then clamps the sum.

magick wrote:Compile with HDRI and you will get the same result for both commands. Without HDRI you get a difference due to clamping the result between 0 and QuantumRange. The -fx option adds R+G+B then clamps. With -color-matrix, it clamps R, G, and B and only then does it sum R+G+B then clamps the sum.

Thanks, I understand. I worked out another way that works without HDRI.


rf=-0.40
gf=1.44
bf=-0.04
convert zelda3.png -color-matrix "\
$rf $gf $bf \
$rf $gf $bf \
$rf $gf $bf \
" \
zelda3_gray_mix3.png

The last method given is the correct one for grayscaling with various weights.

See http://www.imagemagick.org/Usage/color_mods/#grayscale

HDRI is not needed for either FX or Matrix operators.
It is needed for some of the other 'piecemeal' techniques (like separate image adds), to get exactly the same results, especially if weightings are outside the 0..1 range. Basically you are hitting quantum rounding and clipping errors.