[Gdal-dev] Inaccuracy concerning PseudoColor-Tables

[Schmitz, Uwe]

Hi,

I've detected a slight inaccuracy during translation
of a Gtiff formatted file.

I have a file f2.tif on which "tiffinfo -c" gives me:

TIFF Directory at offset 0xf4691e
  Image Width: 4000 Image Length: 4000
  Bits/Sample: 8
  Sample Format: unsigned integer
  Compression Scheme: None
  Photometric Interpretation: palette color (RGB from colormap)
  Samples/Pixel: 1
  Rows/Strip: 2
  Planar Configuration: single image plane
  Color Map: 
       0: 65535 65535 65535
       1: 65535     0     0
       2: 57054 65021 65535
       3: 13107 26214 65535
       4: 61166 53713 49858
       5: 57825 50629 43947
       6: 63736 52428 43176
<<snip>>

After doing

gdal_translate -co COMPRESS=PACKBITS f2.tif f3.tif

a "tiffinfo -c" for f3.tif results in:

TIFF Directory at offset 0x4080a
  Image Width: 4000 Image Length: 4000
  Bits/Sample: 8
  Sample Format: unsigned integer
  Compression Scheme: PackBits
  Photometric Interpretation: palette color (RGB from colormap)
  Samples/Pixel: 1
  Rows/Strip: 2
  Planar Configuration: single image plane
  Color Map: 
       0: 65280 65280 65280
       1: 65280     0     0
       2: 56832 64768 65280
       3: 13056 26112 65280
       4: 60928 53504 49664
       5: 57600 50432 43776
       6: 63488 52224 43008
<<snip>>

Please note that every color value
(except 0) has slightly changed. Well, gdalinfo produces
the same Output for both files, but it works with 8 bit
resolution.

Also using the identify command line tool from
ImageMagick (which also uses 8 bit channel resolution),
I get the following output:


File f2.tif:

f2.tif TIFF 4000x4000 PseudoClass 256c 15.3mb 0.656u 0:01
Image: f2.tif
  Format: TIFF (Tagged Image File Format)
  Geometry: 4000x4000
  Class: PseudoClass
  Type: Palette
  Endianess: LSB
  Colorspace: RGB
  Channel depth:
    Red: 8-bits
    Green: 8-bits
    Blue: 8-bits
  Channel statistics:
    <<channel stats snipped>>
  Colors: 256
         0: (255,255,255)	white
         1: (255,  0,  0)	red
         2: (222,253,255)	#DEFDFF
         3: ( 51,102,255)	#3366FF
         4: (238,209,194)	#EED1C2
         5: (225,197,171)	#E1C5AB
         6: (248,204,168)	#F8CCA8
<<snip>>

File f3.tif:

__3.tif TIFF 4000x4000 PseudoClass 256c 276kb 0.672u 0:01
Image: __3.tif
  Format: TIFF (Tagged Image File Format)
  Geometry: 4000x4000
  Class: PseudoClass
  Type: Palette
  Endianess: LSB
  Colorspace: RGB
  Channel depth:
    Red: 9-bits
    Green: 9-bits
    Blue: 9-bits
  Channel statistics:
    <<channel stats snipped>>
  Colors: 256
         0: (254,254,254)	#FEFEFE
         1: (254,  0,  0)	#FE0000
         2: (221,252,254)	#DDFCFE
         3: ( 50,101,254)	#3265FE
         4: (237,208,193)	#EDD0C1
         5: (224,196,170)	#E0C4AA
         6: (247,203,167)	#F7CBA7
<<snip>>

(Please recognize the unusual channel depth)

I've tested these with FWTools0.9.8 under Windows
and with gdal1.2.6 under Unix. I think this is a bug,
but may be I'm  missing something.

Best wishes
Uwe

---------------------------------------------------------
... Uwe Schmitz  Landesvermessungsamt Nordrhein-Westfalen
... Muffendorfer Str. 19 - 21              D - 53177 Bonn
... E-mail:       uwe.schmitzNO at SPAMlverma.nrw.de
... Internet:     http://www.lverma.nrw.de