[mapserver-dev] Extending LUT syntax to support rasters with bit depth > 8 bit ?

[Even Rouault]

Hi,

Currently if we want to use the LUT with 16 bit raster, we have necessary to 
use SCALE before to reduce the dynamics to 8 bits, since LUT are only 
supported for [0,255]->[0,255] mapping. However, for example in the case of a 
12 bit raster where the desired LUT [0,4095]->[0,255] would have segments with 
a width of input values < 16, it is not possible to derive a proper 8 bit LUT.

For example, let's assume we want this LUT : 0:0,15:15,4080:240,4095:255
and that the input raster dynamics is effectively [0,4095]. Once we have 
rescaled to 8 bits, we cannot "convert" this 12 bit LUT into a 8 bit LUT since 
it would become this degenerated LUT: 0:0,0:15,255:240,255:255

The current pseudo code of LoadGDALImages() is :

(1) if there are no SCALE processing option,
	- acquire the imagery on 8 bit
	- apply 8 bit LUTs when they're defined
(2) otherwise if there's a SCALE processing option
	- acquire the imagery on 32 bit floats
	- do the scaling to get a 8 bit buffer
	- apply 8 bit LUTs when they're defined

My proposal would be to modify (1) to detect first if there's a LUT defined with 
its maximum input value greater than 255. In which case, the imagery would be 
acquired on 16 bit unsigned and the extended LUT would be applied.

It seems to me a natural extension of the current LUT syntax that shouldn't 
have backward compatibility issues.

For now, I would be interested only in 16 bit unsigned imagery, but the same 
idea could potentially be later applied to other cases like floating point 
values.

Opinions ?

Even

-- 
Spatialys - Geospatial professional services
http://www.spatialys.com