102 replies to this topic

[PI_CO100]

Francois was very kind to share the draft Matlab code with me so I tried it on my old 2017 images which were compromised by high clouds . The result included here  is a test and in greyscale and yes I had a problem with the flats back in 2017 and my Moon disc just looks rubbish, however the detail in the corona shows that the alignment process worked really well.

 

I am simply really really excited by the progress made and I cannot wait to see the new developments on this front (the wait is killing me!!). Kudos to Francois and team they are doing something simply AMAZING   

 

Piero

 

 

Attached Thumbnails

• 

[Francois314]

Hi Arnaud,

 

i think the code will be simple enough that you can use it without programing experience, maybe we will make a tutorial before releasing it

 

where are you going in Arpil for the TSE ? I will be near Austin (may be closer from the border if there are clouds)

Edited by Francois314, 09 January 2024 - 11:58 PM.

[Francois314]

HI Piero, it's beautiful, we are making good progress, you should see Jonathan HDR algorithm, and Colin's work on ACHF !

[Francois314]

hi Pachacoti, i sent you an email, just making sure you got it

[Francois314]

here is a quick update and question that we have not solved yet (so any help is appreciated)  

 

the issue is the subpixel alignment.

 

it is linked to what equation we use. the image below summarize the issue. if we use the full convolution as recommended by Druckmüller's we get sharp peaks (usually one for the moon and another for the sun, and maybe one or two more if the data is noisy) on the other hand when we use the FFT convolution without the normalization (as recommended in some online image processing classes) we get a smooth curve much easier to process for the sub pixel interpolation but much harder to know if our alignment is correct (did the moon shift the curve by 1/3 of a pixel? non one knows  )

 

 

let us know what you think !

Edited by Francois314, 10 January 2024 - 01:28 AM.

[PI_CO100]

..you should see Jonathan HDR algorithm, and Colin's work on ACHF !

Anytime I'd  be extremely intrigued! 

[Kranenburg]

Hi Arnaud,

 

i think the code will be simple enough that you can use it without programing experience, maybe we will make a tutorial before releasing it

 

where are you going in Arpil for the TSE ? I will be near Austin (may be closer from the border if there are clouds)

That would be great! I will be in Mabank, Texas. I will fly from Amsterdam to Dallas and have already booked a hotel there. Cloud cover seems to be 50% chance, but I have been lucky so far. This will be my 6th eclipse, all against a clear sky. ;-) Let's hope my luck holds.

[pachacoti]

Francois, I got your email! Regarding your question here, I'm wondering if we can replace the Moon's silhouette with NAN values after edge detection, and then perform Fourier transform by ignoring pixels flagged as NAN? Will this help get rid of, or at least suppress the peak due to the motion of the Moon? Just my 2c

 

 

 

here is a quick update and question that we have not solved yet (so any help is appreciated)  

 

the issue is the subpixel alignment.

 

it is linked to what equation we use. the image below summarize the issue. if we use the full convolution as recommended by Druckmüller's we get sharp peaks (usually one for the moon and another for the sun, and maybe one or two more if the data is noisy) on the other hand when we use the FFT convolution without the normalization (as recommended in some online image processing classes) we get a smooth curve much easier to process for the sub pixel interpolation but much harder to know if our alignment is correct (did the moon shift the curve by 1/3 of a pixel? non one knows  )

 

 

let us know what you think !

[rslobins]

I have been following developments on this thread very closely.

 

I have digital and digitized film image sets of total solar eclipses.   I have digital series starting from 2006 and it includes 2009, 2012, 2015, 2016, 2017 and 2019.   Longer-duration eclipses, if shot from a fixed or alt-azimuth tripod, will show field rotation at high magnification.   This includes SolarQuest mounts, as I have done a fair amount of H-alpha images from August through December 2023 and orienting X-class flares and erupting prominences has been a serious challenge.

 

These would come with calibration frames.   I follow Druckmuller's 2009 MMV instructions.

 

I recall that Dr Druckmuller did a separate stack of the lunar surface images and blended them on top of the smeared moon.

 

 

I was in Texas, based in San Antonio, for the annular eclipse.   I lived in South Texas 1997-2000.

 

Texas was great then, not good now.   Getting around can be real painful.   Do not stop on interstates or highways to observe; people drive fast there, like 80 miles/hour or 130 km/hour fast.

 

Many Americans have moved there for work opportunities and relative freedom than what states like California offer.   Regrettably, and especially in places like Austin with its high-tech and education industry, many there have narcissistic tendencies that manifest on the road.   Driving between San Antonio and Austin is a nightmare, even on Saturday night!   Austin to Dallas is better--it's more real Texas.   The toll roads, although outside the eclipse path, are worth it because time is at a premium.

 

Interstate 35 is a major artery, like your aorta.   If some idiot chooses to have an accident party, that could create massive traffic issues.

 

You need your navigation skills and you ride at night.   Monday morning rush hour will be problematic.

 

Dallas and San Antonio have been growing very fast, and there is a lot of road building in progress.   Imagine driving in a bowl of spaghetti and trying to make a lane change from entering on the right to make a left-exit quickly at high speed.

 

If you are on country highways, that could be better because real Texans are much better drivers.  But don't get lost.   Make sure you have maps or GPS that is UP TO DATE; demand that from your car rental facility!

 

You could scout the area and talk to locals.   Ask permission and treat them very respectfully and they will be very hospitable, helpful and interested.   This applies to most of the areas around and along the path of totality into Ohio.   And if the weather is not cooperating in Texas, you will need 2 days to cover the 1200 miles to Ohio--one way.

 

And make sure that you all eat that barbeque and TexMex food.   There is a top BBQ restaurant in Hondo and there are many very good places all over the area, including Arkansas and Missouri.   West, TX is the center of Czech culture and you can get great Czech food there.

Edited by rslobins, 14 January 2024 - 02:09 AM.

[Francois314]

hi rslobins thanks for your insights, i we would love to see some of the eclipse photographs

 

here is today's update, maybe the most progress we made in a long time.

 

as posted on Jan 10th, it's difficult to know which equation for the FFT convolution we should use. I asked a few people with expertise in the field and they all said the same thing: "it's complicated"

 

Then I realized that the picture Christian gave us were all taken at the same exposure and these pictures aligned perfectly. So i tried to do some exposure compensation.

 

first i tried to use the exposure time

 

exposure_ratio = ImExpo(i) / ImExpo(i+1);

Image_Crop{i} = uint16(double(Image_Crop{i}) / exposure_ratio);

 

that failed miserably (it did not improve the FFT conv)

 

then I tried to use the mean of the images 

 

exposure_ratio = mean2(Image{i}) / mean2(Image{i+1});

Image_Crop{i} = uint16(double(Image_Crop{i}) / exposure_ratio);

 

that was obviously too simple and failed also

 

then i found out that we use the camera response function (CRF) to adjust for exposure (never heard of it, I am only a chemical engineer after all) 

here are two links that  helped me

https://www.mathwork...amresponse.html

https://www.youtube....h?v=95DNdbxaIXE

 

and that worked perfectly, here is a graph of the cross section of two TSE pictures taken at two different exposures (going through the center of the sun horizontally) 

 

before applying the CRF

 

 

After applying the CRF

 

 

it looked like magic and more importantly in cleaned un the FFT convolution nicely

 

 

but does it work ?

[Francois314]

In order to check if the image alignment works, i wrote a simple peace of code that (1) crop the picture around a star (2) interpolate the star in the image using two gaussian functions (one for the rows and one for the columns, i could not find two pictures with round stars out of the 11 data sets i have of the 2017 TSE) (3) find the peak and compute the translation between images

 

I tested it on Piero's data (thanks, you had the cleanest star) and here is how it looks:

 

 

alignment between two images using FFT    4.0   1.0  (no sub pixel yet, need to interpolate the FFT convolution)

alignment between two images using a star  4.5   0.7 

 

looks like it works !

 

it's only tested only on a couple of images. We need to do more testing but i am so exited.

Edited by Francois314, 18 January 2024 - 12:59 AM.

[R Botero]

hi rslobins thanks for your insights, i we would love to see some of the eclipse photographs

here is today's update, maybe the most progress we made in a long time.

as posted on Jan 10th, it's difficult to know which equation for the FFT convolution we should use. I asked a few people with expertise in the field and they all said the same thing: "it's complicated"

Then I realized that the picture Christian gave us were all taken at the same exposure and these pictures aligned perfectly. So i tried to do some exposure compensation.

first i tried to use the exposure time

exposure_ratio = ImExpo(i) / ImExpo(i+1);
Image_Crop{i} = uint16(double(Image_Crop{i}) / exposure_ratio);

that failed miserably (it did not improve the FFT conv)

then I tried to use the mean of the images

exposure_ratio = mean2(Image{i}) / mean2(Image{i+1});
Image_Crop{i} = uint16(double(Image_Crop{i}) / exposure_ratio);

that was obviously too simple and failed also

then i found out that we use the camera response function (CRF) to adjust for exposure (never heard of it, I am only a chemical engineer after all)
here are two links that helped me
https://www.mathwork...amresponse.html
https://www.youtube....h?v=95DNdbxaIXE

and that worked perfectly, here is a graph of the cross section of two TSE pictures taken at two different exposures (going through the center of the sun horizontally)

before applying the CRF



After applying the CRF



it looked like magic and more importantly in cleaned un the FFT convolution nicely



but does it work ?

Eureka!

[Juzwuz]

Also, if you switch to python... FNRGF is implemented in sunkit_image and there are some examples of how to use it:
https://docs.sunpy.o...dial.fnrgf.html
 

I was looking at some Python packages and came across imreg_dft and sunkit_image which do image registration using the phase correlation method. Anybody have experience with these packages? 

 

https://pythonhosted...dft/index.html#

 

https://scikit-image...ter-rotation-py

[valentin_v]

Dear all,

to me it seems that the strong correlation signal coming from the moon should be suppressed when a radial filter is applied to the image as described in "PHASE CORRELATION METHOD FOR THE ALIGNMENT OF TOTAL SOLAR ECLIPSE IMAGES", by the equation (4). Did anyone try to  implement this filter in Matlab?

 

Valentin

[Jhillphotos]

Dear all,

to me it seems that the strong correlation signal coming from the moon should be suppressed when a radial filter is applied to the image as described in "PHASE CORRELATION METHOD FOR THE ALIGNMENT OF TOTAL SOLAR ECLIPSE IMAGES", by the equation (4). Did anyone try to  implement this filter in Matlab?

 

Valentin

Hi Valentin. Thanks for your reply! We have tried implementing this filter and have found both Photoshop and MATLAB to introduce artifacts via a low quality radial blur procedure. A function in DIPlib in Python seems to show promising results though, which is what I'm experimenting with. As Hana stated in her paper, I believe isolating these radial features is a good way to suppress any FFT spectrum corruption from the complex tangential brightness gradient. This would also suppress the effects of the overexposed region of the inner corona in longer exposures. It does remove some of the moon effects, but a window function is still required to suppress spectrum leakage from the image edges and the effects of the moon. 

[ch-viladrich]

Hi guys,

 

On my side, I have the registration step going, based on registration of images using lunar disk (Iris with script) and a short Julia code shifting the images according to the relative solar/lunar motion (ephemerids).

 

But I am having really hard time with the FNRGF of Sunkit package. Has anyone succeeded in using it ? It requires FIT files with WCS data (world coordonnates system), which is not really obvious to populate starting with TIF files taken with a DSLR ...

[Anhydrite]

That would be great! I will be in Mabank, Texas. I will fly from Amsterdam to Dallas and have already booked a hotel there. Cloud cover seems to be 50% chance, but I have been lucky so far. This will be my 6th eclipse, all against a clear sky. ;-) Let's hope my luck holds.

I fly into Austin.  Will possibly be in Kileen at Family members house.....Or wherever the weather is good.

Considering driving into Mexico if I have too.

 

Dragging a lot of gear with me.

[Francois314]

Hi Christian,

 

i am glad you got the alignment working. i'd love to see your image alignment. Very different direction from our work and it sounds interesting. Lets meet with Collin, he implemented the gaussian normalization, last time we met a month ago, his result was amazing, i had never seen so many details in the corona, his results with Phill's data put drukmuller to shame.

 

Sorry i have not been more active, work has been crazy. I am spending the little time I have focusing on camera control, i hope i will be ready for the TSE.

 

Quick question, i never had to check for meridian flip before, but some people mentioned that the meridian flip in Texas is near the time of the eclipse. How do i check ?

 

Best,

 

Francois

[ch-viladrich]

Hi François,

 

I've seen Phil and Collin results :

 

https://www.cloudyni...and-8k-cameras/

 

There are incredible !!

 

Regarding meridien flip, I am lucky, I won't have any in Mexico. More precisely, my mount allows to go beyond the meridien. I set the limit to one hour.

Doing this, and checking the mount /telescope camera position at contacts C1, C2, C3, C4, I can see that I still have a few cm left before the camera touches the tripod.

 

I checked the mount/telescope position by entering the date/hour of the eclipse and coordonates of observation site in the keypad of the mount.

 

I don't know if you can do something similar on your mount ?

[SkipW]

Quick question, i never had to check for meridian flip before, but some people mentioned that the meridian flip in Texas is near the time of the eclipse. How do i check ?

 

Regarding meridien flip, I am lucky, I won't have any in Mexico. More precisely, my mount allows to go beyond the meridien. I set the limit to one hour.

Doing this, and checking the mount /telescope camera position at contacts C1, C2, C3, C4, I can see that I still have a few cm left before the camera touches the tripod.

 

I checked the mount/telescope position by entering the date/hour of the eclipse and coordonates of observation site in the keypad of the mount.

 

I don't know if you can do something similar on your mount ?

This technique is good as far as it goes, but may also depend on the latitude setting of the polar axis.

 

When you tested, if the PA was set for a latitude (France?) that is significantly different than the latitude you will be observing from (Mexico?), you may want to adjust the polar axis to the (presumably) lower latitude and repeat the test.

[knightowl]

Hi, Francois,

 

Very interesting initiative!

 

I have imaged the 2017 eclipse and got this far with photoshop: https://www.flickr.c...157708807181381

 

I will be in the USA for the 2024 eclipse as well.

 

Unfortunately I have no experience with matlab or python, but perhaps a user friendly version could also be produced. I have a lot of raw data files if you would need test images.

 

Clear skies

 

Arnaud

Beautiful images Arnaud!

I'm preparing for the 2024 eclipse, getting my pastels ready. I regret that I didn't make any drawings or notes during the 2017 eclipse and since then I've lost the image in my mind's eye! So for this upcoming one I'm going to devote 3m to sketching so the image is burned into my brain.

Curious--how closely does your image match what's seen with the naked eye? As compared to Druckmuller's? In terms of corona structure, brightness, sky color?

I recall the moon appearing like a pitch-black pupil next to the bright white corona, like an ominous eyeball looking down on me. So cool.

[ch-viladrich]

This technique is good as far as it goes, but may also depend on the latitude setting of the polar axis.

 

When you tested, if the PA was set for a latitude (France?) that is significantly different than the latitude you will be observing from (Mexico?), you may want to adjust the polar axis to the (presumably) lower latitude and repeat the test.

Indeed, I should have mentioned it

Edited by ch-viladrich, 15 March 2024 - 07:27 AM.

[Jhillphotos]

Beautiful images Arnaud!
I'm preparing for the 2024 eclipse, getting my pastels ready. I regret that I didn't make any drawings or notes during the 2017 eclipse and since then I've lost the image in my mind's eye! So for this upcoming one I'm going to devote 3m to sketching so the image is burned into my brain.

Curious--how closely does your image match what's seen with the naked eye? As compared to Druckmuller's? In terms of corona structure, brightness, sky color?
I recall the moon appearing like a pitch-black pupil next to the bright white corona, like an ominous eyeball looking down on me. So cool.

The corona is so challenging to properly display in a way that gives you a similar feeling as the real thing!! The most accurate images I've found for visual representation were taken by Nicolas here: https://hdr-astropho...naked-eye-view/
He gives instructions in his article on how to view the images on your computer screen to produce a similar experience. The first time I did this, it took me right back to 2017! So cool!! I'd definitely give it a shot.

[Kranenburg]

Beautiful images Arnaud!

I'm preparing for the 2024 eclipse, getting my pastels ready. I regret that I didn't make any drawings or notes during the 2017 eclipse and since then I've lost the image in my mind's eye! So for this upcoming one I'm going to devote 3m to sketching so the image is burned into my brain.

Curious--how closely does your image match what's seen with the naked eye? As compared to Druckmuller's? In terms of corona structure, brightness, sky color?

I recall the moon appearing like a pitch-black pupil next to the bright white corona, like an ominous eyeball looking down on me. So cool.

Thank you Jonathan!

 

The visual appearance of the eclipse is quite different, although no features are added or omitted. The moons dark disc is pitch black, I have not been able to see any detail on the disc. Perhaps the biggest difference is the light fall-ff from the lunar disc outward. This is a quadratic fall-off, so in reality the inner part is very bright, a ring-like structure, like you describe. To make larger coronal structures better visible, we try to reduce the light reduction and the ring is lost. This makes my image less 'visual'. Also the images are sharpened to bring out details. Prepare for a very different eclipse in 2024 compared to 2017. The streamers will be far more pronouned. The eclipse of 2017 was during a quit sun, right now the sun is raging with activity. I remember the eclipse of 2001, where the corona was starlike, with many streamers radiating out of the sun. Same for the eclipse of 2016, when the sun was average in activity. I vividly remember seeing one of the streamers in high contrast against the darker sky, something that was less noticable in 2017.

 

I'll see if a can re-process an image stack to bring out a more visual image

[Kranenburg]

Thank you Jonathan!

 

The visual appearance of the eclipse is quite different, although no features are added or omitted. The moons dark disc is pitch black, I have not been able to see any detail on the disc. Perhaps the biggest difference is the light fall-ff from the lunar disc outward. This is a quadratic fall-off, so in reality the inner part is very bright, a ring-like structure, like you describe. To make larger coronal structures better visible, we try to reduce the light reduction and the ring is lost. This makes my image less 'visual'. Also the images are sharpened to bring out details. Prepare for a very different eclipse in 2024 compared to 2017. The streamers will be far more pronouned. The eclipse of 2017 was during a quit sun, right now the sun is raging with activity. I remember the eclipse of 2001, where the corona was starlike, with many streamers radiating out of the sun. Same for the eclipse of 2016, when the sun was average in activity. I vividly remember seeing one of the streamers in high contrast against the darker sky, something that was less noticable in 2017.

 

I'll see if a can re-process an image stack to bring out a more visual image

This would be my best effort for a visual impression. No sharpening was applied. The solar flare is washed out. I found it very difficult to discern any purple color visually, the flares are very bright. I once saw a flare through a maksutov telecope, but in my memory it was more yellowish green than red. A bright flare at the 2016 eclipse gave me a more white-blue impression.

The streamers were weak in the 2017 eclipse and I could hardly see them. As noted, this is expected to be very different in the 2024 eclipse.

 

I had plans to draw my visual impression as well, but I'm afraid I won't have time to run the camera's and draw it as well. I'm very interested in your results!

Attached Thumbnails

•