2

I am using GDAL (gdal_merge) to put together some landsat photos using the proper bands. When I open the GeoTIFF that is created using gdal_merge (in Windows), the GeoTIFF is super dark. Now, when I open the same GeoTIFF in ArcMap, it's nice and bright and realistic looking. I was wondering if there was a function in GDAL that would allow me to do what ArcMap is doing. I have tried using PIL (in python) to enhance the brightness, contrast, and color - and I have had some success, but certainly it's not as nice as what ArcMap is doing.

Does anyone have any ideas? Just in case you are wondering, the gdal_merge command is as follows

gdal_merge.py -separate -ps 16 16 -co PHOTOMETRIC=RGB -o "merged_new.tif" "LC08_L1TP_046029_20180717_20180717_01_RT_b4.tif" "LC08_L1TP_046029_20180717_20180717_01_RT_b3.tif" "LC08_L1TP_046029_20180717_20180717_01_RT_b2.tif"

But I don't think the gdal_merge command is an issue, as I am opening the result of gdal_merge with arcmap and it looks nice.

PolyGeo
  • 65,136
  • 29
  • 109
  • 338
jonesy19
  • 107
  • 1
  • 14
  • 1
    ArcMap automatically applies a stretch to your raster, I think the default for RGB is percent clip, if the resulting image is still detected to be dark a gamma is also applied... look at your symbology tab in ArcMap and note the settings. If you want to see the image like it is in ArcMap in a picture viewer you can export it using renderer to 'brighten up' the image. – Michael Stimson Jul 24 '18 at 23:06

1 Answers1

5

ArcMap stretches the values of a raster automatically by some clip algorithm (percent clip).

You can use gdal_translate -scale option to "scale"/stretch the values.

-scale [src_min src_max [dst_min dst_max]]:
Rescale the input pixels values from the range src_min to src_max to the range dst_min to dst_max. If omitted the output range is 0 to 255. If omitted the input range is automatically computed from the source data.

So you need to find out the right values by yourself. Use gdalinfo -hist merged_new.tif to find see the values of the file.

Assuming most values (98%) lie between values of 20 and 150, we can do this:

gdal_translate -of GTiff -ot Byte -scale 20 150 0 255 merged_new.tif merged_new_scaled.tif

For your Landsat scene, you posted this information:

Metadata: STATISTICS_MAXIMUM=35050 STATISTICS_MEAN=3832.5215314015 STATISTICS_MINIMUM=0 STATISTICS_STDDEV=4860.9555695374 Band 3 Block=9697x1 Type=UInt16, ColorInterp=Blue

That is a 16 Bit image. If you want an 8bit (=Byte) image to view it in an image viewer, you need to scale down the values with something like this -ot Byte -scale 0 65535 0 255 to keep all values, or you can cut at the maximum: -ot Byte -scale 0 35050 0 255.

But that will probably still be too dark, as your mean is 3832 and stddev is 4860, so you might even want -ot Byte -scale 0 16384 0 255. As you will then cut some high values, you will lose information in very bright spots like clouds and glaciers.

Related:

pLumo
  • 6,489
  • 20
  • 41
  • Thanks, I will try this out and get back to the thread with results. Is that true that 98% of values lie between 20 and 150 (as a general rule of thumb)? – jonesy19 Jul 26 '18 at 03:40
  • No not at all.That is individual – pLumo Jul 26 '18 at 04:41
  • Hmm...So using the command you referenced above, it creates a new tif, but the tif file is completely white.
    I ran the gdalinfo -hist command, and it gave me three sets of answers (I'm guessing one for each colorband). There's a lot of info on the output, however. What values from this output are the values that you are referencing in your answer? Should I be looking here? Metadata: STATISTICS_MAXIMUM=35050 STATISTICS_MEAN=3832.5215314015 STATISTICS_MINIMUM=0 STATISTICS_STDDEV=4860.9555695374 Band 3 Block=9697x1 Type=UInt16, ColorInterp=Blue Thanks!     – jonesy19 Jul 26 '18 at 17:35
  • What do you mean by the tif file is completely white ? The tif file has values not colors, the program you open the file with "decides" which value is which color. You need to scale the values to mach a "normal" 8 bit image, I updated the answer. – pLumo Jul 27 '18 at 07:18
  • 1
    Thanks so much for the updated answer. Although your numbers gave me a completely black picture, it put me on the right track.

    I used the following value "-ot Byte -scale 0 150 0 255", and it gave me a nice bright image. You don't have to respond, as I've got what I need, but out of curiosity - do you know why that number worked? If I go with a higher number, it starts to get darker and darker.

    Thanks!

     – jonesy19 Aug 02 '18 at 19:45
  • Also, do you know if there is a way to make the image more vibrant with color? I think I have the -scale parameter down pretty well, but it looks fairly washed out now. Thanks. – jonesy19 Aug 02 '18 at 23:41
  • look into the -exponent option of gdal_translate. But you might be better off using an Image Editor ;-) – pLumo Aug 03 '18 at 05:24
  • Yeah, unfortunately, I need to do this via scripts. The old one (using arcpy) seems to do it pretty easily. I definitely can brighten up the image (using your suggestion with scale, as well as using the python PIL module), but the colors still seem washed out. I will look into -exponent. Thanks. :) – jonesy19 Aug 03 '18 at 16:02
  • Can you expand on how/why a standard deviation of 4860 means a max scaling value of 16384 is appropriate? – matt wilkie Jul 12 '21 at 18:11
  • @pLumo your exponent hint led me to wonder https://gis.stackexchange.com/questions/403776/what-is-the-gdal-translate-scale-exponent-option-for – matt wilkie Jul 12 '21 at 18:46