[GRASS-user] Color equalization after patching/mosaicking the satellite data from multiple scenes

GRASS GRASS user
1

Hi all

I have some data from different scenes of Landsat8 covering my area of interest. But after mosaicking, it clearly shows the border line and the difference in colors in the composites.

For example I am attaching two cases, one from Sentinel2 and another from Landsat8.

I mosaicked two scenes of sentinel taken on the same day,- November 19 2016, and tried following commands to get a composite.

i.colors.enhance blue=S2A_B02 green=S2A_B8A.hpf red=S2A_B11.hpf
r.composite blue=S2A_B02.hpf green=S2A_B8A.hpf red=S2A_B11 output=test_118A02

And it works perfect, there is no sign that the image is from two scenes (attached - S2_19Nov2016comp_118A02.jpg ).

But in the case of L8, it is more complicated as the data is coming from three scenes. Two scenes are from the same orbit and third from adjacent which has a different date.
For each scene, there are two acquisitions in November 2016 (Total six images in November 2016 for this g.region).
Hence I used r.patch and r.series to create a November month composite and with this, the color composite looks really bad.

i.colors.enhance blue=L8_B02.hpf green=L8_B05.hpf red=L8_B06.hpf
r.composite blue=L8_B02.hpf green=L8_B05.hpf red=L8_B06.hpf output=test_652

Please see attached “L8_Nov2016_comp_652.jpg”

Is there anything I miss here to obtain a good composite over multiple scenes taken from different dates, but from the same month?

Regards

Sajid

(attachments)

L8_Nov2016_comp_652.jpg
S2_19Nov2016comp_118A02.jpg

2

* Sajid Pareeth <spareeth@gmail.com> [2016-12-30 16:34:23 +0100]:

Hi all

I have some data from different scenes of Landsat8 covering my area of
interest. But after mosaicking, it clearly shows the border line and the
difference in colors in the composites.

Dear Sajid,

some normalisation, before patching, is what "needs" to be done, I think. Relative
normalisation here (either for images of different dates over the same
area, or neighbouring images of the same date).

I never finished implementing some very simple process as described in
user manual(s) for QuickBird2 imagery, if I remember well.

I have a draft module which I named i.radio.balance. In short: the
major difference between two scenes of the same area is the solar
geometry. This difference can be minimized by correcting imagery for
Earth-sun distance and solar zenith angle.

I can send you off-list, if you want, the on-going script (incomplete, yet well
documented I believe it to be).

Cheers, Nikos

ps- We need more advanced algorithms for such tasks (ie relative
radiomatric normalisation).

For example I am attaching two cases, one from *Sentinel2* and another from
*Landsat8*.

I mosaicked two scenes of sentinel taken on the *same day,- November 19
2016*, and tried following commands to get a composite.

i.colors.enhance blue=S2A_B02 green=S2A_B8A.hpf red=S2A_B11.hpf

r.composite blue=S2A_B02.hpf green=S2A_B8A.hpf red=S2A_B11
output=test_118A02

And it works perfect, there is no sign that the image is from two scenes
(attached - S2_19Nov2016comp_118A02.jpg ).

But in the case of L8, it is more complicated as the data is coming from
three scenes. Two scenes are from the same orbit and third from adjacent
which has a different date.
For each scene, there are two acquisitions in November 2016 (Total six
images in November 2016 for this g.region).
Hence I used r.patch and r.series to create a November month composite and
with this, the color composite looks really bad.

i.colors.enhance blue=L8_B02.hpf green=L8_B05.hpf red=L8_B06.hpf

r.composite blue=L8_B02.hpf green=L8_B05.hpf red=L8_B06.hpf output=test_652

Please see attached "L8_Nov2016_comp_652.jpg"

Is there anything I miss here to obtain a good composite over multiple
scenes taken from different dates, but from the same month?

Regards

Sajid

_______________________________________________
grass-user mailing list
grass-user@lists.osgeo.org
http://lists.osgeo.org/mailman/listinfo/grass-user

--
Nikos Alexandris | Remote Sensing & Geomatics
GPG Key Fingerprint 6F9D4506F3CA28380974D31A9053534B693C4FB3

3

Hi Nikos,

Thank you for the reply.

···

some normalisation, before patching, is what “needs” to be done, I think. Relative
normalisation here (either for images of different dates over the same
area, or neighbouring images of the same date).

Yes I also agree with that, I tried doing histogram matching before running r.patch, though I am not sure if it is right way. Any idea on how the change in values due to matching will affect further processing?

I never finished implementing some very simple process as described in
user manual(s) for QuickBird2 imagery, if I remember well.

I have a draft module which I named i.radio.balance. In short: the
major difference between two scenes of the same area is the solar
geometry. This difference can be minimized by correcting imagery for
Earth-sun distance and solar zenith angle.

I can send you off-list, if you want, the on-going script (incomplete, yet well
documented I believe it to be).

I would be happy to test your code and let you know the results. Let me know.

Regards

Sajid