Lab written as txt:

Post by **snibgo** » 2013-08-15T22:17:23-07:00

There seems to be a bug writing txt: format from Lab images, in IM versions dating back to v6.8.0.

IM writes and reads integers with an offset of Quantum/2, but negative values are written as zero. Thus writing a Lab txt file and reading it back in will yield different values.

When reading txt files, any values < 0 or > 32767 are treated as negative, and this is reasonable.

Test cases (IM v6.8.6-0, Windows):

Text file lab_test_neg contains:

Code: Select all

# ImageMagick pixel enumeration: 10,1,65535,cielab
0,0: (59712,    0,        0)
1,0: (59712,    10,       0)
2,0: (59712,    100,      0)
3,0: (59712,    1000,     0)
4,0: (59712,    -10,      0)
5,0: (59712,    -100,     0)
6,0: (59712,    -1000,    0)
7,0: (59712,    32767,    0)
8,0: (59712,    32768,    0)
9,0: (59712,    0,        0)

Reading test:

Code: Select all

%I%convert ^
  txt:lab_test_neg.txt ^
  -set colorspace Lab ^
  txt:

Result of reading test:

Code: Select all

D:\web\im>c:\im\ImageMagick-6.8.6-Q16fix\convert   txt:lab_test_neg.txt   -set colorspace Lab txt:
# ImageMagick pixel enumeration: 10,1,65535,cielab
0,0: (59712,    0,    0)  #E94000000000  cielab(91.1147%,0%,0%)
1,0: (59712,   10,    0)  #E940000A0000  cielab(91.1147%,0.015259%,0%)
2,0: (59712,  100,    0)  #E94000640000  cielab(91.1147%,0.15259%,0%)
3,0: (59712, 1000,    0)  #E94003E80000  cielab(91.1147%,1.5259%,0%)
4,0: (59712,    0,    0)  #E94000000000  cielab(91.1147%,-0.015259%,0%)
5,0: (59712,    0,    0)  #E94000000000  cielab(91.1147%,-0.15259%,0%)
6,0: (59712,    0,    0)  #E94000000000  cielab(91.1147%,-1.5259%,0%)
7,0: (59712,32767,    0)  #E9407FFF0000  cielab(91.1147%,49.9992%,0%)
8,0: (59712,    0,    0)  #E94000000000  cielab(91.1147%,-50.0008%,0%)
9,0: (59712,    0,    0)  #E94000000000  cielab(91.1147%,0%,0%)

The cielab percentages are good, so I think IM correctly reads Lab txt files.

Writing test:

Code: Select all

%I%convert ^
  -size 1x10 ^
  gradient:rgb(0%%,100%%,100%%)-rgb(100%%,100%%,0%%) ^
  -colorspace Lab ^
  -write txt: ^
  -colorspace sRGB ^
  txt:

Result:

Code: Select all

D:\web\im>c:\im\ImageMagick-6.8.6-Q16fix\convert   -size 1x10   gradient:rgb(0%,100%,100%)-rgb(100%,100%,0%)   -colorspace Lab   -write txt:   -colorspace sRGB txt:
# ImageMagick pixel enumeration: 1,10,65535,cielab
0,0: (59712,    0,    0)  #E94000000000  cielab(91.1147%,-18.8556%,-5.54665%)
0,1: (59282,    0,    0)  #E79200000000  cielab(90.4585%,-21.5946%,-0.158694%)
0,2: (59032,    0, 3529)  #E69800000DC9  cielab(90.0771%,-23.5309%,5.38491%)
0,3: (58996,    0, 7242)  #E67400001C4A  cielab(90.0221%,-24.4221%,11.0506%)
0,4: (59190,    0,10978)  #E73600002AE2  cielab(90.3182%,-24.1459%,16.7514%)
0,5: (59623,    0,14625)  #E8E700003921  cielab(90.9789%,-22.6978%,22.3163%)
0,6: (60296,    0,18005)  #EB8800004655  cielab(92.0058%,-20.1923%,27.4739%)
0,7: (61203,    0,20870)  #EF1300005186  cielab(93.3898%,-16.8276%,31.8456%)
0,8: (62330,    0,22971)  #F37A000059BB  cielab(95.1095%,-12.8344%,35.0515%)
0,9: (63658,    0,24281)  #F8AA00005ED9  cielab(97.1359%,-8.45197%,37.0504%)
# ImageMagick pixel enumeration: 1,10,65535,srgb
0,0: (    0,65535,65534)  #0000FFFFFFFE  srgb(0%,100%,99.9985%)
0,1: ( 7291,65535,58254)  #1C7BFFFFE38E  srgb(11.1254%,100%,88.8899%)
0,2: (14565,65535,50972)  #38E5FFFFC71C  srgb(22.2248%,100%,77.7783%)
0,3: (21845,65535,43691)  #5555FFFFAAAB  srgb(33.3333%,100%,66.6682%)
0,4: (29126,65535,36408)  #71C6FFFF8E38  srgb(44.4434%,100%,55.555%)
0,5: (36407,65535,29127)  #8E37FFFF71C7  srgb(55.5535%,100%,44.445%)
0,6: (43690,65535,21846)  #AAAAFFFF5556  srgb(66.6667%,100%,33.3349%)
0,7: (50972,65535,14563)  #C71CFFFF38E3  srgb(77.7783%,100%,22.2217%)
0,8: (58254,65535, 7284)  #E38EFFFF1C74  srgb(88.8899%,100%,11.1147%)
0,9: (65534,65535,    2)  #FFFEFFFF0002  srgb(99.9985%,100%,0.0030518%)

The sRGB values are more or less correct.

The cielab percentages are more or less correct, remembering that mid-values in a and b channels are 0%. But all negative a or b percentages are shown with integer zero.

So the bug I'd like fixing is that negative integers when written shouldn't be clipped to zero.

(The Luv colorspace isn't offset by Quantum/2, so never goes negative and doesn't have this problem.)

Post by **magick** » 2013-08-16T09:28:03-07:00

Here are the before and after results for HDRI sRGB => Lab => sRGB:

Code: Select all

# ImageMagick pixel enumeration: 1,10,65535,srgb
0,0: (    0,65535,65535)  #0000FFFFFFFF  cyan
0,1: ( 7282,65535,58253)  #1C72FFFFE38D  srgb(11.1111%,100%,88.8889%)
0,2: (14563,65535,50972)  #38E3FFFFC71C  srgb(22.2222%,100%,77.7778%)
0,3: (21845,65535,43690)  #5555FFFFAAAA  srgb(85,255,170)
0,4: (29127,65535,36408)  #71C7FFFF8E38  srgb(44.4444%,100%,55.5556%)
0,5: (36408,65535,29127)  #8E38FFFF71C7  srgb(55.5556%,100%,44.4444%)
0,6: (43690,65535,21845)  #AAAAFFFF5555  srgb(170,255,85)
0,7: (50972,65535,14563)  #C71CFFFF38E3  srgb(77.7778%,100%,22.2222%)
0,8: (58253,65535, 7282)  #E38DFFFF1C72  srgb(88.8889%,100%,11.1111%)
0,9: (65535,65535,    0)  #FFFFFFFF0000  yellow
# ImageMagick pixel enumeration: 1,10,65535,cielab
0,0: (59712,    0,    0)  #E94000000000  cielab(91.1152%,-18.8549%,-5.54738%)
0,1: (59282,    0,    0)  #E79200000000  cielab(90.4578%,-21.5952%,-0.158643%)
0,2: (59032,    0, 3529)  #E69800000DC9  cielab(90.077%,-23.5312%,5.38474%)
0,3: (58996,    0, 7242)  #E67400001C4A  cielab(90.0218%,-24.4224%,11.0513%)
0,4: (59190,    0,10978)  #E73600002AE2  cielab(90.3182%,-24.1455%,16.7512%)
0,5: (59623,    0,14625)  #E8E700003921  cielab(90.9793%,-22.6976%,22.3169%)
0,6: (60296,    0,18005)  #EB8800004655  cielab(92.0062%,-20.1927%,27.4745%)
0,7: (61203,    0,20870)  #EF1300005186  cielab(93.3894%,-16.8274%,31.8452%)
0,8: (62330,    0,22971)  #F37A000059BB  cielab(95.1096%,-12.8347%,35.052%)
0,9: (63658,    0,24281)  #F8AA00005ED9  cielab(97.1365%,-8.45184%,37.0509%)
# ImageMagick pixel enumeration: 1,10,65535,srgb
0,0: (    0,65535,65535)  #0000FFFFFFFF  srgb(-5.98795e-05%,100%,100%)
0,1: ( 7282,65535,58253)  #1C72FFFFE38D  srgb(11.1111%,100%,88.8889%)
0,2: (14563,65535,50972)  #38E3FFFFC71C  srgb(22.2222%,100%,77.7778%)
0,3: (21845,65535,43690)  #5555FFFFAAAA  srgb(33.3333%,100%,66.6667%)
0,4: (29127,65535,36408)  #71C7FFFF8E38  srgb(44.4444%,100%,55.5556%)
0,5: (36408,65535,29127)  #8E38FFFF71C7  srgb(55.5556%,100%,44.4444%)
0,6: (43690,65535,21845)  #AAAAFFFF5555  srgb(66.6667%,100%,33.3333%)
0,7: (50972,65535,14563)  #C71CFFFF38E3  srgb(77.7778%,100%,22.2222%)
0,8: (58253,65535, 7282)  #E38DFFFF1C72  srgb(88.8889%,100%,11.1111%)
0,9: (65535,65535,    0)  #FFFFFFFF0000  srgb(100%,100%,-4.11654e-05%)

Now the Q16 unsigned short results:

Code: Select all

# ImageMagick pixel enumeration: 1,10,65535,srgb
0,0: (    0,65535,65535)  #0000FFFFFFFF  cyan
0,1: ( 7282,65535,58253)  #1C72FFFFE38D  srgb(11.1116%,100%,88.8884%)
0,2: (14563,65535,50972)  #38E3FFFFC71C  srgb(22.2217%,100%,77.7783%)
0,3: (21845,65535,43690)  #5555FFFFAAAA  srgb(85,255,170)
0,4: (29127,65535,36408)  #71C7FFFF8E38  srgb(44.445%,100%,55.555%)
0,5: (36408,65535,29127)  #8E38FFFF71C7  srgb(55.555%,100%,44.445%)
0,6: (43690,65535,21845)  #AAAAFFFF5555  srgb(170,255,85)
0,7: (50972,65535,14563)  #C71CFFFF38E3  srgb(77.7783%,100%,22.2217%)
0,8: (58253,65535, 7282)  #E38DFFFF1C72  srgb(88.8884%,100%,11.1116%)
0,9: (65535,65535,    0)  #FFFFFFFF0000  yellow
# ImageMagick pixel enumeration: 1,10,65535,cielab
0,0: (59712,    0,    0)  #E94000000000  cielab(91.1147%,-18.8556%,-5.54665%)
0,1: (59282,    0,    0)  #E79200000000  cielab(90.4585%,-21.5946%,-0.158694%)
0,2: (59032,    0, 3529)  #E69800000DC9  cielab(90.0771%,-23.5309%,5.38491%)
0,3: (58996,    0, 7242)  #E67400001C4A  cielab(90.0221%,-24.4221%,11.0506%)
0,4: (59190,    0,10978)  #E73600002AE2  cielab(90.3182%,-24.1459%,16.7514%)
0,5: (59623,    0,14625)  #E8E700003921  cielab(90.9789%,-22.6978%,22.3163%)
0,6: (60296,    0,18005)  #EB8800004655  cielab(92.0058%,-20.1923%,27.4739%)
0,7: (61203,    0,20870)  #EF1300005186  cielab(93.3898%,-16.8276%,31.8456%)
0,8: (62330,    0,22971)  #F37A000059BB  cielab(95.1095%,-12.8344%,35.0515%)
0,9: (63658,    0,24281)  #F8AA00005ED9  cielab(97.1359%,-8.45197%,37.0504%)
# ImageMagick pixel enumeration: 1,10,65535,srgb
0,0: (    0,65535,65534)  #0000FFFFFFFE  srgb(0%,100%,99.9985%)
0,1: ( 7291,65535,58254)  #1C7BFFFFE38E  srgb(11.1254%,100%,88.8899%)
0,2: (14565,65535,50972)  #38E5FFFFC71C  srgb(22.2248%,100%,77.7783%)
0,3: (21845,65535,43691)  #5555FFFFAAAB  srgb(33.3333%,100%,66.6682%)
0,4: (29126,65535,36408)  #71C6FFFF8E38  srgb(44.4434%,100%,55.555%)
0,5: (36407,65535,29127)  #8E37FFFF71C7  srgb(55.5535%,100%,44.445%)
0,6: (43690,65535,21846)  #AAAAFFFF5556  srgb(66.6667%,100%,33.3349%)
0,7: (50972,65535,14563)  #C71CFFFF38E3  srgb(77.7783%,100%,22.2217%)
0,8: (58254,65535, 7284)  #E38EFFFF1C74  srgb(88.8899%,100%,11.1147%)
0,9: (65534,65535,    2)  #FFFEFFFF0002  srgb(99.9985%,100%,0.0030518%)

Note, some Lab values do not map directly into the sRGB colorspace. Given the above, exactly which values do you think are incorrect and what value should they be?

Post by **snibgo** » 2013-08-16T12:38:59-07:00

Yes, that shows the problem. I have reddened the incorrect values, for just the Q16 version. The corresponding hex values are also wrong.

# ImageMagick pixel enumeration: 1,10,65535,srgb
0,0: ( 0,65535,65535) #0000FFFFFFFF cyan
0,1: ( 7282,65535,58253) #1C72FFFFE38D srgb(11.1116%,100%,88.8884%)
0,2: (14563,65535,50972) #38E3FFFFC71C srgb(22.2217%,100%,77.7783%)
0,3: (21845,65535,43690) #5555FFFFAAAA srgb(85,255,170)
0,4: (29127,65535,36408) #71C7FFFF8E38 srgb(44.445%,100%,55.555%)
0,5: (36408,65535,29127) #8E38FFFF71C7 srgb(55.555%,100%,44.445%)
0,6: (43690,65535,21845) #AAAAFFFF5555 srgb(170,255,85)
0,7: (50972,65535,14563) #C71CFFFF38E3 srgb(77.7783%,100%,22.2217%)
0,8: (58253,65535, 7282) #E38DFFFF1C72 srgb(88.8884%,100%,11.1116%)
0,9: (65535,65535, 0) #FFFFFFFF0000 yellow
# ImageMagick pixel enumeration: 1,10,65535,cielab
0,0: (59712, 0, 0) #E94000000000 cielab(91.1147%,-18.8556%,-5.54665%)
0,1: (59282, 0, 0) #E79200000000 cielab(90.4585%,-21.5946%,-0.158694%)
0,2: (59032, 0, 3529) #E69800000DC9 cielab(90.0771%,-23.5309%,5.38491%)
0,3: (58996, 0, 7242) #E67400001C4A cielab(90.0221%,-24.4221%,11.0506%)
0,4: (59190, 0,10978) #E73600002AE2 cielab(90.3182%,-24.1459%,16.7514%)
0,5: (59623, 0,14625) #E8E700003921 cielab(90.9789%,-22.6978%,22.3163%)
0,6: (60296, 0,18005) #EB8800004655 cielab(92.0058%,-20.1923%,27.4739%)
0,7: (61203, 0,20870) #EF1300005186 cielab(93.3898%,-16.8276%,31.8456%)
0,8: (62330, 0,22971) #F37A000059BB cielab(95.1095%,-12.8344%,35.0515%)
0,9: (63658, 0,24281) #F8AA00005ED9 cielab(97.1359%,-8.45197%,37.0504%)
# ImageMagick pixel enumeration: 1,10,65535,srgb
0,0: ( 0,65535,65534) #0000FFFFFFFE srgb(0%,100%,99.9985%)
0,1: ( 7291,65535,58254) #1C7BFFFFE38E srgb(11.1254%,100%,88.8899%)
0,2: (14565,65535,50972) #38E5FFFFC71C srgb(22.2248%,100%,77.7783%)
0,3: (21845,65535,43691) #5555FFFFAAAB srgb(33.3333%,100%,66.6682%)
0,4: (29126,65535,36408) #71C6FFFF8E38 srgb(44.4434%,100%,55.555%)
0,5: (36407,65535,29127) #8E37FFFF71C7 srgb(55.5535%,100%,44.445%)
0,6: (43690,65535,21846) #AAAAFFFF5556 srgb(66.6667%,100%,33.3349%)
0,7: (50972,65535,14563) #C71CFFFF38E3 srgb(77.7783%,100%,22.2217%)
0,8: (58254,65535, 7284) #E38EFFFF1C74 srgb(88.8899%,100%,11.1147%)
0,9: (65534,65535, 2) #FFFEFFFF0002 srgb(99.9985%,100%,0.0030518%)

The formula for the integer cielab values should be something like
int_value = percent_value /100 * 65535

So the correct integer values will be (shown in green, but I haven't calculated the correct hex values):

# ImageMagick pixel enumeration: 1,10,65535,srgb
0,0: ( 0,65535,65535) #0000FFFFFFFF cyan
0,1: ( 7282,65535,58253) #1C72FFFFE38D srgb(11.1116%,100%,88.8884%)
0,2: (14563,65535,50972) #38E3FFFFC71C srgb(22.2217%,100%,77.7783%)
0,3: (21845,65535,43690) #5555FFFFAAAA srgb(85,255,170)
0,4: (29127,65535,36408) #71C7FFFF8E38 srgb(44.445%,100%,55.555%)
0,5: (36408,65535,29127) #8E38FFFF71C7 srgb(55.555%,100%,44.445%)
0,6: (43690,65535,21845) #AAAAFFFF5555 srgb(170,255,85)
0,7: (50972,65535,14563) #C71CFFFF38E3 srgb(77.7783%,100%,22.2217%)
0,8: (58253,65535, 7282) #E38DFFFF1C72 srgb(88.8884%,100%,11.1116%)
0,9: (65535,65535, 0) #FFFFFFFF0000 yellow
# ImageMagick pixel enumeration: 1,10,65535,cielab
0,0: (59712,-12357, -3635) #E940???????? cielab(91.1147%,-18.8556%,-5.54665%)
0,1: (59282,-14152, -104) #E792???????? cielab(90.4585%,-21.5946%,-0.158694%)
0,2: (59032,-15421, 3529) #E698????0DC9 cielab(90.0771%,-23.5309%,5.38491%)
0,3: (58996,-16005, 7242) #E674????1C4A cielab(90.0221%,-24.4221%,11.0506%)
0,4: (59190,-15824,10978) #E736????2AE2 cielab(90.3182%,-24.1459%,16.7514%)
0,5: (59623,-14875,14625) #E8E7????3921 cielab(90.9789%,-22.6978%,22.3163%)
0,6: (60296,-13233,18005) #EB88????4655 cielab(92.0058%,-20.1923%,27.4739%)
0,7: (61203,-11028,20870) #EF13????5186 cielab(93.3898%,-16.8276%,31.8456%)
0,8: (62330,-7756,22971) #F37A????59BB cielab(95.1095%,-12.8344%,35.0515%)
0,9: (63658,-5539,24281) #F8AA????5ED9 cielab(97.1359%,-8.45197%,37.0504%)
# ImageMagick pixel enumeration: 1,10,65535,srgb
0,0: ( 0,65535,65534) #0000FFFFFFFE srgb(0%,100%,99.9985%)
0,1: ( 7291,65535,58254) #1C7BFFFFE38E srgb(11.1254%,100%,88.8899%)
0,2: (14565,65535,50972) #38E5FFFFC71C srgb(22.2248%,100%,77.7783%)
0,3: (21845,65535,43691) #5555FFFFAAAB srgb(33.3333%,100%,66.6682%)
0,4: (29126,65535,36408) #71C6FFFF8E38 srgb(44.4434%,100%,55.555%)
0,5: (36407,65535,29127) #8E37FFFF71C7 srgb(55.5535%,100%,44.445%)
0,6: (43690,65535,21846) #AAAAFFFF5556 srgb(66.6667%,100%,33.3349%)
0,7: (50972,65535,14563) #C71CFFFF38E3 srgb(77.7783%,100%,22.2217%)
0,8: (58254,65535, 7284) #E38EFFFF1C74 srgb(88.8899%,100%,11.1147%)
0,9: (65534,65535, 2) #FFFEFFFF0002 srgb(99.9985%,100%,0.0030518%)

The same principle applies to the HDRI results.

Thanks.

Post by **snibgo** » 2013-08-16T13:56:18-07:00

Another simple round-trip test RGB -> Lab-> txt_file -> RGB:

Code: Select all

%I%convert ^
  -size 1x10 ^
  gradient:rgb(0%%,100%%,100%%)-rgb(100%%,100%%,0%%) ^
  -write rgb0.png ^
  -colorspace Lab ^
  -write txt:lab_text.txt ^
  +delete ^
  lab_text.txt ^
  -set colorspace Lab ^
  -colorspace sRGB ^
  rgb1.png

%IM%compare -metric RMSE rgb0.png rgb1.png NULL:

The comparison result should be (very near) zero. It was in v6.7.9 but not in subsequent versions.

Post by **magick** » 2013-08-16T17:09:14-07:00

We can reproduce the problem you posted and have a patch in ImageMagick 6.8.6-9 Beta available by sometime tomorrow. Thanks.

Post by **snibgo** » 2013-08-16T17:42:13-07:00

Excellent. Thanks.

Legacy ImageMagick Discussions Archive

Lab written as txt:

Lab written as txt:

Re: Lab written as txt:

Re: Lab written as txt:

Re: Lab written as txt:

Re: Lab written as txt:

Re: Lab written as txt: