bleeding star fix?
- Thread starter shurik
- Start date
georg.viehoever
Well-known member
Hi,
- you can upload images to the forum by using the "Additional Options..." button when posting something.
- PI currently does not have a de-blooming filter. I suppose it is not high on the priority list of missing modules because of the philospophy of PI never to "invent" data where there is none. For example, if there are faint stars hidden behind the blooming streaks, the common filters cannot recover them.
- A technique proposed in http://pixinsight.com/forum/index.php?topic=3426.msg23577#msg23577 is to rotate your sensor when taking pictures, and replacing the bloomed data by real data using pixel rejection during image integration.
- use the search button in the forum to find more on "blooming".
Georg
- you can upload images to the forum by using the "Additional Options..." button when posting something.
- PI currently does not have a de-blooming filter. I suppose it is not high on the priority list of missing modules because of the philospophy of PI never to "invent" data where there is none. For example, if there are faint stars hidden behind the blooming streaks, the common filters cannot recover them.
- A technique proposed in http://pixinsight.com/forum/index.php?topic=3426.msg23577#msg23577 is to rotate your sensor when taking pictures, and replacing the bloomed data by real data using pixel rejection during image integration.
- use the search button in the forum to find more on "blooming".
Georg
Carlos Milovic
Well-known member
Do HDR 
. Take shorter exposures, and replace the saturated area with the info of the shorter expositions.
. Take shorter exposures, and replace the saturated area with the info of the shorter expositions.
Carlos Milovic said:Do HDR
There are only two correct ways to deal with bloomings, which Georg and Carlos have pointed out here. Bloomings are essentially a dynamic range problem, so the best way to deal with them is by integrating short (unsaturated) exposures. HDRCombination is the tool of choice.
As noted by Georg, cosmetic deblooming tools can only invent data to replace saturated areas. We don't favor these 'solutions', and hence we have decided not to invest resources in a deblooming tool. This is not a matter of 'religion' as I've read somewhere, but a matter of doing things correctly.
WOW, I'm absolutely shocked by this comment!
You make an imaging TOOL specifically designed to PERTURB astronomical data into something we like to see (COMPRESSED DYNAMIC RANGE) but you draw the line at de-blooming stars???
And yet I have seen a beta-debloomer on the forum some posts back or perhaps years back!!! What's going on?
If it a matter of time or priorities then please send me the beta script and I'll try to completely it.
Best regards
Neil.
You make an imaging TOOL specifically designed to PERTURB astronomical data into something we like to see (COMPRESSED DYNAMIC RANGE) but you draw the line at de-blooming stars???
And yet I have seen a beta-debloomer on the forum some posts back or perhaps years back!!! What's going on?
If it a matter of time or priorities then please send me the beta script and I'll try to completely it.
Best regards
Neil.
Juan Conejero said:Carlos Milovic said:Do HDR
There are only two correct ways to deal with bloomings, which Georg and Carlos have pointed out here. Bloomings are essentially a dynamic range problem, so the best way to deal with them is by integrating short (unsaturated) exposures. HDRCombination is the tool of choice.
As noted by Georg, cosmetic deblooming tools can only invent data to replace saturated areas. We don't favor these 'solutions', and hence we have decided not to invest resources in a deblooming tool. This is not a matter of 'religion' as I've read somewhere, but a matter of doing things correctly.
Agreed to a point - try taking an image of B33 with Alnitak in the frame with a true linear grade ccd such as the ST10?
What exposure do you propose?
The difference in apparent brightness is enormous between the nebula and Alnitak and no detector can come close to the dynamic range required, right!
NABG ccd camera simply drain off the excess charge which the result that the ccd has a none linear response etc...
Regards
Neil
Hi Neil,
Firstly, I disagree with the idea that dynamic range compression perturbs astronomical data. Dynamic range compression refers to a set of image processing algorithms and techniques with the purpose of remapping image data to make them representable on devices with limited dynamics, such as computer monitors and digital printers, or to adequate them to the characteristics and response of the human vision system. DRC techniques do not create, remove or modify existing data in arbitrary ways; they apply transformations in a purely algorithmic and repeatable way and are based on sound principles that respect the documentary contents of the data.
A cosmetic deblooming tool replaces CCD blooming artifacts with adjacent or guessed data (for example, using inpainting algorithms), and basically works in the same way as an automatic clone stamp tool. While the clone stamp is perfectly admissible to repair little cosmetic defects smaller than the PSF of the image, it should not be used to replace relatively large image areas. Arbitrary replacement of data can be admissible in purely artistic photography, but not in astrophotography. The key concept here is inventing data. For example, what happens if a blooming artifact is covering one or more stars? Our policy regarding bloomings can be summarized in two points:
- If possible, use observational techniques and HDR composition to replace bloomings with real data.
- If the above is not feasible or unavailable, leave the bloomings intact. Here is an example of nice bloomings in a nice image, and here is another case where bloomings were completely fixed with observational techniques.
We explored some ideas in 2010 to develop a deblooming tool using 'smart' inpainting algorithms. We dropped the project because it requires a substantial R&D investment and does not provide, in our opinion, a valid solution to the blooming problem. We have a lot of work to do in much more interesting areas.
We didn't go much beyond initial exploration steps, so I can't provide anything usable as a basis for further development. The basic idea is to inpaint saturated pixels using diffusion algorithms such as GREYCstoration. You can see an example here. The main problem of GREYCstoration is that it doesn't work with linear data.
Firstly, I disagree with the idea that dynamic range compression perturbs astronomical data. Dynamic range compression refers to a set of image processing algorithms and techniques with the purpose of remapping image data to make them representable on devices with limited dynamics, such as computer monitors and digital printers, or to adequate them to the characteristics and response of the human vision system. DRC techniques do not create, remove or modify existing data in arbitrary ways; they apply transformations in a purely algorithmic and repeatable way and are based on sound principles that respect the documentary contents of the data.
A cosmetic deblooming tool replaces CCD blooming artifacts with adjacent or guessed data (for example, using inpainting algorithms), and basically works in the same way as an automatic clone stamp tool. While the clone stamp is perfectly admissible to repair little cosmetic defects smaller than the PSF of the image, it should not be used to replace relatively large image areas. Arbitrary replacement of data can be admissible in purely artistic photography, but not in astrophotography. The key concept here is inventing data. For example, what happens if a blooming artifact is covering one or more stars? Our policy regarding bloomings can be summarized in two points:
- If possible, use observational techniques and HDR composition to replace bloomings with real data.
- If the above is not feasible or unavailable, leave the bloomings intact. Here is an example of nice bloomings in a nice image, and here is another case where bloomings were completely fixed with observational techniques.
We explored some ideas in 2010 to develop a deblooming tool using 'smart' inpainting algorithms. We dropped the project because it requires a substantial R&D investment and does not provide, in our opinion, a valid solution to the blooming problem. We have a lot of work to do in much more interesting areas.
We didn't go much beyond initial exploration steps, so I can't provide anything usable as a basis for further development. The basic idea is to inpaint saturated pixels using diffusion algorithms such as GREYCstoration. You can see an example here. The main problem of GREYCstoration is that it doesn't work with linear data.
I'm not going to start a flame war of words - its just my stand point that once you apply image processing to astronomical data you are in fact performing a "one-way" none-reversal processes to make an acceptable image for the human eye, no matter what system you choose to use to finally display it. Of course, keeping the original FIT's intact maintains the integrity of that data.
From a marketing point of view, there are two other commercial imaging programs in the market that offer de-blooming tools; MaxIm DL and CCDware. There may well be others; however, I have chosen PixInsight for the abundance of dedicated tools for astronomical image manipulation. That being said there is a huge apparent oversight, the de-blooming tool?
Creating a deblooming tool to help your users to debloom their images in the best possible manner, even if you consider it to be a compromised solution would be appreciated. Let your users worry about the ethics of inventing data, that?s not really your concern?
Or if you still consider this to be unethical then create a MS style dialog box asking "Are you really sure that you want to do this?"
I'm trying to be fair an not blow this out of all proportions but it does seem that you are trying to enforce some sort of "image processing ethical standard" on your users!
This being the case, please explain to me your stand point on the "MorphologicalTransformation" tool This definatey should come with such an MS style dialog box right?
Best regards
Neil.
From a marketing point of view, there are two other commercial imaging programs in the market that offer de-blooming tools; MaxIm DL and CCDware. There may well be others; however, I have chosen PixInsight for the abundance of dedicated tools for astronomical image manipulation. That being said there is a huge apparent oversight, the de-blooming tool?
Creating a deblooming tool to help your users to debloom their images in the best possible manner, even if you consider it to be a compromised solution would be appreciated. Let your users worry about the ethics of inventing data, that?s not really your concern?
Or if you still consider this to be unethical then create a MS style dialog box asking "Are you really sure that you want to do this?"
I'm trying to be fair an not blow this out of all proportions but it does seem that you are trying to enforce some sort of "image processing ethical standard" on your users!
This being the case, please explain to me your stand point on the "MorphologicalTransformation" tool
This definatey should come with such an MS style dialog box right?Best regards
Neil.
I'm not going to start a flame war of words - its just my stand point that once you apply image processing to astronomical data you are in fact performing a "one-way" none-reversal processes to make an acceptable image for the human eye, no matter what system you choose to use to finally display it. Of course, keeping the original FIT's intact maintains the integrity of that data.
From a marketing point of view, there are two other commercial imaging programs in the market that offer de-blooming tools; MaxIm DL and CCDware. There may well be others; however, I have chosen PixInsight for the abundance of dedicated tools for astronomical image manipulation. That being said there is a huge apparent oversight, the de-blooming tool?
Creating a deblooming tool to help your users to debloom their images in the best possible manner, even if you consider it to be a compromised solution would be appreciated. Let your users worry about the ethics of inventing data, that?s not really your concern?
Or if you still consider this to be unethical then create a MS style dialog box asking "Are you really sure that you want to do this?"
I'm trying to be fair and not blow this out of all proportions but it does seem that you are trying to enforce some sort of "image processing ethical standard" on your users!
This being the case, please explain to me your stand point on the "MorphologicalTransformation" tool This definitely should come with such an MS style dialog box right?
I checked the proposed other solutions and I don't want to rotate the ccd camera in the middle of an imaging session, this was the only viable alternative. The bird example was interesting.
Best regards
Neil.
From a marketing point of view, there are two other commercial imaging programs in the market that offer de-blooming tools; MaxIm DL and CCDware. There may well be others; however, I have chosen PixInsight for the abundance of dedicated tools for astronomical image manipulation. That being said there is a huge apparent oversight, the de-blooming tool?
Creating a deblooming tool to help your users to debloom their images in the best possible manner, even if you consider it to be a compromised solution would be appreciated. Let your users worry about the ethics of inventing data, that?s not really your concern?
Or if you still consider this to be unethical then create a MS style dialog box asking "Are you really sure that you want to do this?"
I'm trying to be fair and not blow this out of all proportions but it does seem that you are trying to enforce some sort of "image processing ethical standard" on your users!
This being the case, please explain to me your stand point on the "MorphologicalTransformation" tool
This definitely should come with such an MS style dialog box right?I checked the proposed other solutions and I don't want to rotate the ccd camera in the middle of an imaging session, this was the only viable alternative. The bird example was interesting.
Best regards
Neil.
First of all, what's the problem with making short exposures (and, if necessary, really short ones) and combining them to fix your bloomings with the HDRComposition tool? This technique works extremely well and requires just a bit of extra acquisition and processing work.
The way you process your images is important. It is in fact as important as---and in my opinion even more important than---the final result. For example, hand painted masks and selective sharpening based on arbitrary manual selections are unacceptable IMO; they don't belong to astrophotography as I understand it. We already have discussed this topic many times on this forum. If necessary we can resort to old threads, or we can start a new discussion here if you want, or both, but my point is that not everything is valid in astrophotography.
In my opinion we have huger omissions than a deblooming tool. We have many exciting tools and algorithms to work in the short-middle term. Some examples:
- Further development of wavelet-median transforms
- Curvelets
- Noise reduction with anisotropic diffusion and adaptive Gaussian filtering
- Wavelet regularized maximum entropy deconvolution
- A layering system
- Improved version of StarAlignment tolerant of distortion
- Featureless image registration for panorama and mosaic construction
- Applications of fuzzy sets
- Applications of image segmentation with snakes
- Applications of principal component analysis
- Applications of computational geometry
Along with these, this year I want to start working on GPU acceleration with OpenCL.
It isn't that I don't want to work on a deblooming tool, it's just that we have a lot of priorities that are much more important, IMHO, than a cosmetic debloomer. Regarding marketing, well, who know me can tell that I am not a marketing person by any stretch of the imagination. Otherwise I probably wouldn't be here working on PixInsight years ago.
That said, if anyone wants to start a deblooming tool as a new development project, I'll support her or him as well as I can.
Not at all. In general, I am against enforcing anything. However, I try to be an ethical person, and image processing and astrophotography are very important for me. So when I have to apply priorities to our development work---and I have to apply them all the time because our resources are very limited---they usually reflect my own points of view and interests, sometimes more than others. The same happens when somebody asks me about critical topics such as the present one: I tend to express my opinions loud and clear. As I said before, I am not a marketing kind of person.
Mathematical morphology. Wikipedia has a good general description on the topic (when it is available again after the SOPA/PIPA blacking out).
As I said at the beginning of this post, the best way to fix bloomings is with the HDRComposition tool and a range of short exposures. Why isn't this a viable alternative for you?
The way you process your images is important. It is in fact as important as---and in my opinion even more important than---the final result. For example, hand painted masks and selective sharpening based on arbitrary manual selections are unacceptable IMO; they don't belong to astrophotography as I understand it. We already have discussed this topic many times on this forum. If necessary we can resort to old threads, or we can start a new discussion here if you want, or both, but my point is that not everything is valid in astrophotography.
In my opinion we have huger omissions than a deblooming tool. We have many exciting tools and algorithms to work in the short-middle term. Some examples:
- Further development of wavelet-median transforms
- Curvelets
- Noise reduction with anisotropic diffusion and adaptive Gaussian filtering
- Wavelet regularized maximum entropy deconvolution
- A layering system
- Improved version of StarAlignment tolerant of distortion
- Featureless image registration for panorama and mosaic construction
- Applications of fuzzy sets
- Applications of image segmentation with snakes
- Applications of principal component analysis
- Applications of computational geometry
Along with these, this year I want to start working on GPU acceleration with OpenCL.
It isn't that I don't want to work on a deblooming tool, it's just that we have a lot of priorities that are much more important, IMHO, than a cosmetic debloomer. Regarding marketing, well, who know me can tell that I am not a marketing person by any stretch of the imagination. Otherwise I probably wouldn't be here working on PixInsight years ago.
That said, if anyone wants to start a deblooming tool as a new development project, I'll support her or him as well as I can.
Not at all. In general, I am against enforcing anything. However, I try to be an ethical person, and image processing and astrophotography are very important for me. So when I have to apply priorities to our development work---and I have to apply them all the time because our resources are very limited---they usually reflect my own points of view and interests, sometimes more than others. The same happens when somebody asks me about critical topics such as the present one: I tend to express my opinions loud and clear. As I said before, I am not a marketing kind of person.
Mathematical morphology. Wikipedia has a good general description on the topic (when it is available again after the SOPA/PIPA blacking out).
As I said at the beginning of this post, the best way to fix bloomings is with the HDRComposition tool and a range of short exposures. Why isn't this a viable alternative for you?
I can agree with you on most of the points that you have made accept want does or does not constitute a perturbation of the original data, so lets put that behind us. I do agree with your point regarding hand painted masks etc... so there is also no need to re-post old information; however, here I will make the point that the deconvolution routines are not good enough and I'm also guilty of the following procedure;
1. Make a deconvolution of the original image
2. Mask all
3. Paint thought the mask the improved detail
4. Combine / flatten the final image
And I'm not the only one doing this. The only way forward here is to improve the masking / the selectivity of the deconvolution effects in the first place. and yes I have read your tutorials here!
Unfortunately, the examples here don't help; there is a tendency to use "fanstastic" image data as the starting point which is not what most of us have! Most of us amateur astronomers are trying to get the best possible results from less than Ideal data, that's a fact.
I have no objection to trying the HDRComposition tool and taking shorter exposure. The issue is that I have data now that needs to be corrected, and I live in Northern Europe and might have to wait another week or two before I can try the suggested technique. We don't all have pristine skies and imaging opportunities every week!!!
The frustrating part for me at least is that you have already shown in this forum examples of a deblooming routine that in my opinion looks to be achieving better results than MaxIm DL5's deblooming routines, which as I have said are heavy handed.
From your list of other priorities, I am disappointed to see a "Layering System" at position 5 if these are in order from highest to lowest etc... This is the one omission that ensure the continued use of Photoshop.
1. Make a deconvolution of the original image
2. Mask all
3. Paint thought the mask the improved detail
4. Combine / flatten the final image
And I'm not the only one doing this. The only way forward here is to improve the masking / the selectivity of the deconvolution effects in the first place. and yes I have read your tutorials here!
Unfortunately, the examples here don't help; there is a tendency to use "fanstastic" image data as the starting point which is not what most of us have! Most of us amateur astronomers are trying to get the best possible results from less than Ideal data, that's a fact.
I have no objection to trying the HDRComposition tool and taking shorter exposure. The issue is that I have data now that needs to be corrected, and I live in Northern Europe and might have to wait another week or two before I can try the suggested technique. We don't all have pristine skies and imaging opportunities every week!!!
The frustrating part for me at least is that you have already shown in this forum examples of a deblooming routine that in my opinion looks to be achieving better results than MaxIm DL5's deblooming routines, which as I have said are heavy handed.
From your list of other priorities, I am disappointed to see a "Layering System" at position 5 if these are in order from highest to lowest etc... This is the one omission that ensure the continued use of Photoshop.
zvrastil
Well-known member
Hi Neil,
just a suggestion what I'd try to do in your shoes in attempt to debloom manually with PixInsight:
1) Use RangeSelection to create mask covering saturated & nearly saturated pixels (image A)
2) Create suitable StarMask (image B)
3) Subtract B from A using PixelMath - the tricky part is to set up StarMask parameters so saturated star centers are removed from image A, leaving only bloom spikes
4) Select pixel math result as mask for your image
5) Try to apply MorphologicalTransformation, Erosion with horizontal kernel (assuming bloom spikes are vertical)
6) If needed, parameters in points 1-3 to create better mask
Of course this can't be compared with sophisticated inpainting techniques mentioned by Juan and you may as well use another application to do deblooming.
regards, Zbynek
just a suggestion what I'd try to do in your shoes in attempt to debloom manually with PixInsight:
1) Use RangeSelection to create mask covering saturated & nearly saturated pixels (image A)
2) Create suitable StarMask (image B)
3) Subtract B from A using PixelMath - the tricky part is to set up StarMask parameters so saturated star centers are removed from image A, leaving only bloom spikes
4) Select pixel math result as mask for your image
5) Try to apply MorphologicalTransformation, Erosion with horizontal kernel (assuming bloom spikes are vertical)
6) If needed, parameters in points 1-3 to create better mask
Of course this can't be compared with sophisticated inpainting techniques mentioned by Juan and you may as well use another application to do deblooming.
regards, Zbynek
Nocturnal said:
Layers will make PI more competitive against rival products. As I've said before I am not a marketing guy, but we have to sell licenses, and I am responsible for that. Common wisdom says that layers are necessary in astrophotography. That's false, of course, and is just a 'legacy' of many years of hegemony of applications that have nothing better to offer. The fact is that lack of layers is currently one of the main arguments against PixInsight.
Instead of fighting against the layers concept, I think it is much more productive to adapt it to PixInsight. On the other hand, the layering paradigm isn't a bad idea for some tasks. With a good design and integration, layers can be very useful in PixInsight. We are currently considering the possibility to implement a layering system in the PI Core application, which would allow us to achieve a better integration with the whole platform.
zvrastil said:
My initial idea is along the same line but also a little different as follows;
1. Find the centroid of the top 50 or so brightest stars in the image
2. Of those stars find the ones with blooms
3. Remove the bloom and protect the centroid and a certain size / diameter determined by the none bloomed width???
Now the dilemma is what or how should I replace the bloomed data??? Use the statistical average from around the bloom???
Comments are of course welcome? The inpainting techniques look nice but I couldn't find an formula to implement? So I guess its trial and error time!!!
Regards
Neil.
georg.viehoever
Well-known member
Look at the software pointed to in http://pixinsight.com/forum/index.php?topic=1300.msg25999#msg25999.
One method they offer is to replace bleeded pixels by their counterparts that are 90 degrees away, with the star as a center.
One method they offer is to replace bleeded pixels by their counterparts that are 90 degrees away, with the star as a center.