Skip to content

Commit

Permalink
VRT: add a new mode to apply chained processing steps that apply to s…
Browse files Browse the repository at this point in the history
…everal bands at the same time

The following built-in algorithms are introduced, and typically applied
in the following order:
- Dehazing: remove haze effects by applying (subsampled) gain and offset
  auxiliary datasets.
- BandAffineCombination: to perform an affine transformation combination
  of bands.
- Trimming: local thresholding of saturation
- LUT: apply a look-up table (band per band)
  • Loading branch information
rouault committed Mar 6, 2024
1 parent a354d38 commit d219efd
Show file tree
Hide file tree
Showing 12 changed files with 5,089 additions and 150 deletions.
1,258 changes: 1,258 additions & 0 deletions autotest/gdrivers/vrtprocesseddataset.py

Large diffs are not rendered by default.

13 changes: 13 additions & 0 deletions doc/source/drivers/raster/vrt.rst
Original file line number Diff line number Diff line change
Expand Up @@ -1907,6 +1907,19 @@ See the dedicated :ref:`vrt_multidimensional` page.

vrt_multidimensional

Processed dataset VRT
---------------------

.. versionadded:: 3.9

See the dedicated :ref:`vrt_processed_dataset` page.

.. toctree::
:maxdepth: 1
:hidden:

vrt_processed_dataset

vrt:// connection string
------------------------

Expand Down
261 changes: 261 additions & 0 deletions doc/source/drivers/raster/vrt_processed_dataset.rst
Original file line number Diff line number Diff line change
@@ -0,0 +1,261 @@
.. _vrt_processed_dataset:

================================================================================
VRT processed dataset
================================================================================

.. versionadded:: 3.9

VRT processed dataset is a specific variant of the :ref:`raster.vrt` format,
to apply chained processing steps that apply to several bands at the same time.

The following built-in algorithms are introduced, and may typically be applied
in the following order:

- Dehazing: remove haze effects by applying (subsampled) gain and offset
auxiliary datasets.

- BandAffineCombination: perform an affine transformation combination of bands.

- Trimming: apply local thresholding of saturation

- LUT: apply a look-up table (band per band)

More algorithms can be registered are run-time with the :cpp:func:`GDALVRTRegisterProcessedDatasetFunc`
function`

Here's an example of such a file to apply various correction to a R,G,B,NIR dataset:

.. code-block:: xml
<VRTDataset subClass="VRTProcessedDataset">
<Input>
<SourceFilename relativeToVRT="1">source.tif</SourceFilename>
</Input>
<ProcessingSteps>
<Step name="Dehazing">
<Algorithm>Dehazing</Algorithm>
<Argument name="relative_filenames">true</Argument>
<Argument name="gain_dataset_filename_1">gains.tif</Argument>
<Argument name="gain_dataset_filename_2">gains.tif</Argument>
<Argument name="gain_dataset_filename_3">gains.tif</Argument>
<Argument name="gain_dataset_filename_4">gains.tif</Argument>
<Argument name="gain_dataset_band_1">1</Argument>
<Argument name="gain_dataset_band_2">2</Argument>
<Argument name="gain_dataset_band_3">3</Argument>
<Argument name="gain_dataset_band_4">4</Argument>
<Argument name="offset_dataset_filename_1">offsets.tif</Argument>
<Argument name="offset_dataset_filename_2">offsets.tif</Argument>
<Argument name="offset_dataset_filename_3">offsets.tif</Argument>
<Argument name="offset_dataset_filename_4">offsets.tif</Argument>
<Argument name="offset_dataset_band_1">1</Argument>
<Argument name="offset_dataset_band_2">2</Argument>
<Argument name="offset_dataset_band_3">3</Argument>
<Argument name="offset_dataset_band_4">4</Argument>
<Argument name="nodata">0</Argument>
<Argument name="min">1</Argument>
<Argument name="max">10000</Argument>
</Step>
<Step name="Linear combination">
<Algorithm>BandAffineCombination</Algorithm>
<Argument name="coefficients_1">0,1.2,-0.2,0.0,0.0</Argument>
<Argument name="coefficients_2">0,-0.03,1.03,0.0,0.0</Argument>
<Argument name="coefficients_3">0,0.0,0.0,1.0,0.0</Argument>
<Argument name="coefficients_4">0,0.0,0.0,0.0,1.0</Argument>
<Argument name="min">1</Argument>
<Argument name="max">10000</Argument>
</Step>
<Step name="Trimming">
<Algorithm>Trimming</Algorithm>
<Argument name="relative_filenames">true</Argument>
<Argument name="trimming_dataset_filename">trimming.tif</Argument>
<Argument name="tone_ceil">10000</Argument>
<Argument name="top_margin">0</Argument>
<Argument name="top_rgb">10000</Argument>
</Step>
<Step name="LUT">
<Algorithm>LUT</Algorithm>
<Argument name="lut_1">
0:0,10000.0:255
</Argument>
<Argument name="lut_2">
0:0,10000.0:255
</Argument>
<Argument name="lut_3">
0:0,10000.0:255
</Argument>
<Argument name="lut_4">
0:0,10000.0:255
</Argument>
</Step>
</ProcessingSteps>
<VRTRasterBand dataType="Byte" band="1" subClass="VRTProcessedRasterBand">
<ColorInterp>Red</ColorInterp>
</VRTRasterBand>
<VRTRasterBand dataType="Byte" band="2" subClass="VRTProcessedRasterBand">
<ColorInterp>Green</ColorInterp>
</VRTRasterBand>
<VRTRasterBand dataType="Byte" band="3" subClass="VRTProcessedRasterBand">
<ColorInterp>Blue</ColorInterp>
</VRTRasterBand>
<VRTRasterBand dataType="Byte" band="4" subClass="VRTProcessedRasterBand">
</VRTRasterBand>
</VRTDataset>
.vrt format
-----------

The ``VRTDataset`` root element must have a ``subClass="VRTProcessedDataset"`` attribute.

The following child elements of ``VRTDataset`` may be defined: ``SRS``, ``GeoTransform``, ``Metadata``. If they are not explicitly set, they are inferred from the input dataset.

``VRTRasterBand`` elements may be explicitly defined, in particular if the data type of the virtual dataset after all processing steps is different from the input one, or if the number of output bands is different from the number of input bands. If there is no explicit ``VRTRasterBand`` element, the number and data types of input bands are used implicitly. When explicitly defined, ``VRTRasterBand`` elements must have a ``subClass="VRTProcessedRasterBand"`` attribute.
`
It must also have the 2 following child elements:

- ``Input``, which must have one and only one of the following ``SourceFilename`` or ``VRTDataset`` as child elements, to define the input dataset to which to apply the processing steps.

- ``ProcessingSteps``, with at least one child ``Step`` element.

Each ``Step`` must have a ``Algorithm`` child element, and an optional ``name`` attribute.
The value of ``Algorithm`` must be a registered VRTProcessedDataset function. At time of writing, the following 4 algorithms are defined: ``Dehazing``, ``BandAffineCombination``, ``Trimming`` and ``LUT``.

A ``Step`` will generally have one or several ``Argument`` child elements, some of them being required, others optional. Consult the documentation of each algorithm.

Dehazing algorithm
------------------

Remove haze effects by applying (subsampled) gain and offset auxiliary datasets.

The gain and offset auxiliary datasets must have a georeferencing consistent of
the input dataset, but may have a different resolution.

The formula applied by that algorithm is: ``output_value = clamp(input_value * gain - offset, min, max)``

The following required arguments must be specified:

- ``gain_dataset_filename_{band}``: Filename to the gain dataset, where {band} must be replaced by 1 to the number of input bands.

- ``gain_dataset_band_{band}``: Band number corresponding to ``gain_dataset_filename_{band}``, where {band} must be replaced by 1 to the number of input bands.

- ``offset_dataset_filename_{band}``: Filename to the offset dataset, where {band} must be replaced by 1 to the number of input bands.

- ``offset_dataset_band_{band}``: Band number corresponding to ``offset_dataset_filename_{band}``, where {band} must be replaced by 1 to the number of input bands.


The following optional arguments may be specified:

- ``relative_filenames``: Whether gain and offset filenames are relative to the VRT. Allowed values are ``true`` and ``false``. Defaults to ``false``

- ``min``: Clamp minimum value, applied before writing the output value.

- ``max``: Clamp maximum value, applied before writing the output value.

- ``nodata``: Override the input nodata value coming from the previous step (or the input dataset for the first step).

- ``gain_nodata``: Override the nodata value coming from the gain dataset(s).

- ``offset_nodata``: Override the nodata value coming from the offset dataset(s).


BandAffineCombination algorithm
-------------------------------

Perform an affine transformation combination of bands.

The following required argument must be specified:

- ``coefficients_{band}``: Comma-separated coefficients for combining bands where {band} must be replaced by 1 to the number of output bands. The number of coefficients in each argument must be 1 + number_of_input_bands, where the first coefficient is a constant, the second coefficient is the weight of the first input band, the third coefficient is the weight of the second input band, etc.


The following optional arguments may be specified:

- ``src_nodata``: Override the input nodata value coming from the previous step (or the input dataset for the first step).

- ``dst_nodata``: Set the output nodata value.

- ``replacement_nodata``: Value to substitute to a valid computed value that would be equal to dst_nodata.

- ``dst_intended_datatype``: Intended datatype of output (which might be different than the working data type). Used to infer an appropriate value for replacement_nodata when it is not specified.

- ``min``: Clamp minimum value, applied before writing the output value.

- ``max``: Clamp maximum value, applied before writing the output value.


Trimming algorithm
------------------

Apply local thresholding of saturation, with a special processing of the R,G,B bands compared to other bands.

The pseudo algorithm used for each pixel is:

.. code-block::
// Extract local saturation value from trimming image
localMaxRGB = value from TrimmingImage
reducedRGB = min ( (1-top_margin)*top_rgb/localMaxRGB ; 1)
// RGB bands specific process
RGB[] = get red, green, blue components of input buffer
maxRGB = max(RGB[])
toneMaxRGB = min ( toneCeil/maxRGB ; 1)
toneBand[] = min ( toneCeil/RGB[] ; 1)
output_value_RGB[] = min ( reducedRGB*RGB[]*toneBand[] / toneMaxRGB ; topRGB)
// Other bands processing (NIR, ...): only apply RGB reduction factor
Trimmed(OtherBands[]) = reducedRGB * OtherBands[]
The following required arguments must be specified:

- ``trimming_dataset_filename``: Filename of the trimming dataset. It must have one single band. It must have a georeferencing consistent of the input dataset, but may have a different resolution.

- ``top_rgb``: Maximum saturating RGB output value.

- ``tone_ceil``: Maximum threshold beyond which we give up saturation.

- ``top_margin``: Margin to allow for dynamics in brighest areas (between 0 and 1, should be close to 0)


The following optional arguments may be specified:

- ``relative_filenames``: Whether the trimming dataset filename is relative to the VRT. Allowed values are ``true`` and ``false``. Defaults to ``false``

- ``red_band``: Index (one-based) of the red band. Defaults to 1.

- ``green_band``: Index (one-based) of the green band. Defaults to 1.

- ``blue_band``: Index (one-based) of the blue band. Defaults to 1.

- ``nodata``: Override the input nodata value coming from the previous step (or the input dataset for the first step).

- ``trimming_nodata``: Override the nodata value coming from the trimming dataset.


LUT
---

Apply a look-up table (band per band), typically to get from UInt16 to Byte data types.

The following required argument must be specified:

- ``lut_{band}``: List of the form ``[src value 1]:[dest value 1],[src value 2]:[dest value 2],....``. {band} must be replaced by 1 to the number of bands.


The following optional arguments may be specified:

- ``src_nodata``: Override the input nodata value coming from the previous step (or the input dataset for the first step).

- ``dst_nodata``: Set the output nodata value.
2 changes: 2 additions & 0 deletions frmts/vrt/CMakeLists.txt
Original file line number Diff line number Diff line change
Expand Up @@ -14,6 +14,8 @@ add_gdal_driver(
vrtdataset.cpp
pixelfunctions.cpp
vrtpansharpened.cpp
vrtprocesseddataset.cpp
vrtprocesseddatasetfunctions.cpp
vrtmultidim.cpp
gdaltileindexdataset.cpp
STRONG_CXX_WFLAGS)
Expand Down
Loading

0 comments on commit d219efd

Please sign in to comment.