[gdal-dev] Fwd: gdal_translate histogram conversion

[Andrew C Aitchison]

On Wed, 1 Jun 2022, Pierre Chaponniere wrote:

> Hello gdal community !
>
> I'd like to submit your group a question in regards to the use of
> gdal_translate function.
> I am converting tif images (64bits, 6 bands multispectral) to jpg images
> uint8

uint8 can only have 256 distinct values, but from the top plot it 
appears that you have more than 256 values in each band.

I am not very familiar with the details of the jpeg format, but
many jpeg applications do not support >8bit (eg 12,14, 16 or even more 
bits) values. Some jpeg libraries can support 12 or 16 bit values,
*but only when compiled specially* for such values, and may not support 
8bit images at the same time.

I also note that bande 6 has (all?most) values >30000, which suggests that
they may actually actually be *negative* 16bit values.

> via the Qgis interface to operate in lot but I face issues with the
> conversion of the histogram values.
> It appears this gdal-translate truncate the histogram to the 255 first
> values.
> I therefore tried the scaling (-scale) argument using this line
> -scale 0 65536 0 255
>
> But it appears the resulting histogram is spiky (bottom plot)....compare to
> the original histogram (top plot)
> See image here
> https://ibb.co/mGZHCR0
>
> Any idea from your side on how to better represent histogram from this
> gdal-translate conversion ?

As hinted above,
you will first need to choose a format with more bits than uint8 jpeg.

-- 
Andrew C. Aitchison					Kendal, UK
 			andrew at aitchison.me.uk