This program normalizes the radiometric values of multi-sensor and multi-temporal overhead image time-series.
This is the implementation for the ISPRS 2020 paper: Relative radiometric normalization using several automatically chosen reference images for multi-sensor, multi-temporal series
Hessel, C., Grompone von Gioi, R., Morel, J. M., Facciolo, G., Arias, P., and de Franchis, C.: RELATIVE RADIOMETRIC NORMALIZATION USING SEVERAL AUTOMATICALLY CHOSEN REFERENCE IMAGES FOR MULTI-SENSOR, MULTI-TEMPORAL SERIES, ISPRS Ann. Photogramm. Remote Sens. Spatial Inf. Sci., V-2-2020, 845–852, https://doi.org/10.5194/isprs-annals-V-2-2020-845-2020, 2020.
Version 0.0.1, released on August 20, 2020.
By Charles Hessel, CMLA, ENS Paris-Saclay.
The code contains contributions from Rafael Grompone von Gioi, Tristan Dagobert, Carlo de Franchis.
Main source code repository for this software: https://github.com/chlsl/rrn-multisensor-multidate
The simplest way to install rrn is to run
pip install .from the directory in which the setup.py file is located.
This will compile the two C modules that are included in the project, install
rrn as a package available in your python projects, and also make the
command-line interface rrn available from anywhere on your machine.
This project uses libtiff, libpng and libjpeg.
The required python packages will be installed by pip. They are listed in the
requirements.txt file.
Execute the script tests/test.sh to verify your installation. This should
create a directory out, which contains the output normalized time series.
The command
rrn sequence-registered/*{B02,B02,B04}.tif --band B02 B03 B04 --outdir sequence-registered-stabilizedwill take all images given in argument and save their stabilized version in the
directory "sequence-registered-stabilized". The stabilized images are floating
point tif RGB, and to their names is appended _stab. Tone-mapped versions of
the same images are saved too; they are suffixed _stab_tonemap and saved as
png. The input images can be either RGB or separate bands. Their type must be
specified with the argument --band or -b.
The script takes as input a registered time series.
For RNN to correctly load the input images, their name must contain indications concerning the band and the satellite (and correction level for S2).
_BBB, where BBB stands
for B02, B03, B04 or RGB. The B02 band is supposed to be the blue, B03
the green and B04 the red. You can also give a 3-channels image with the
3 bands already merged, in which case the input names must contain _RGB.
Rename your images if needed._S2A_ and _S2B_; and _L1C_ or
_L2A_ for the correction level._L8_.Example:
The image 2019-07-03_S2B_orbit_094_tile_31UDQ_L2A_band_B02.tif is detected as
a Sentinel-2 image, correction level 2A and band B02 (blue).
You can see more examples in the test directory.
RRN uses several key images so as to handle seasonal color variations.
To obtain more stable sequences, use the --local-max option.
Setting it to a higher value will cause RRN to use fewer key images and
therefore to allow less color change. If you set it to the length of the input
sequence or above, only one key image will be used.
To remove the overlay with the date that appears on the bottom of the stabilized
images, use the flag --no-overlay.
Use rrn --help to get help and a list of available options (reported below).
Relative Radiometric Normalization
optional arguments:
-h, --help show this help message and exit
positional arguments (mandatory):
imlist list of images
mandatory arguments:
-b {RGB,B08,B03,B04,B02} [{RGB,B08,B03,B04,B02} ...], --band {RGB,B08,B03,B04,B02} [{RGB,B08,B03,B04,B02} ...]
space-separated list of bands to load (in the provided
files, see "imlist"). Specifying RGB implies that the
B02, B03 and B04 bands were merged beforehand. Default
is RGB.
-o OUTDIR, --outdir OUTDIR
output directory (created if nonexistent)
parameters of the method:
-m LOCAL_MAX, --local-max LOCAL_MAX
a notion of how "local" the local maxima are. Default
is 9.
--no-visible-ground-area NO_VISIBLE_GROUND_AREA
threshold on the proportion (percentage) of not visible
ground pixels. Images whose proportion of not visible
ground pixels is above this threshold are rejected.
Default is 25.
by-pass some of the method's steps:
--key-images KEY_IMAGES [KEY_IMAGES ...]
list of index for key images.
-c CLOUDS [CLOUDS ...], --clouds CLOUDS [CLOUDS ...]
input cloud masks. If not provided, they are computed
using R. Grompone von Gioi ground visibility detector.
--load-previous use this flag to load previously computed intermediary
results rather than computing them. Useful for
debugging mainly.
alternatives for some steps:
--acoefs-method {ransac,L1constrained,L1,median}
computation method for the affine coefficients. Default
is ransac.
--usecorrsift flag to replace corrsift (slow) by the angles (faster).
The angle error is less accurate. Default is True.
--dont-usecorrsift set to False the usecorrsift flag.
more options!:
--verbose print non-critical information
--no-overlay use this flag to remove the bottom overlay with the
date added on the stabilized images.
--roughnorm flag to apply a rough normalization to the input
images. Default is True.
--no-roughnorm Set to False the roughnorm flag.
--tonemap flag to apply a tonemapping step on the output images.
Default is True. (use --no-tonemap to deactivate)
--no-tonemap set to False the tonemap flag.This program is free software: you can redistribute it and/or modify it under the terms of the GNU Affero General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version.
This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Affero General Public License for more details.
You should have received a copy of the GNU Affero General Public License along with this program. If not, see http://www.gnu.org/licenses/.