Manual Image Integration vs. BPPscript?

[footbag]

I've read in the past that manual image integration for the calibrated lights is better then using BatchPreprocessingScript.  BPP works best for calibration frames.  Is that still the case?  Or have a lot of the features been merged into the newer versions of BPP.

BPP seems to have a lot of settings for image integration itself (including sigma sliders) and I'm not quite sure what settings to change in ImageIintegration to test it out.  I use average combination method with WSC most of the time. 

Are the sigma sliders with pixel rejection images the best things change when trying to improve an image? 

[cdesselles]

I keep wondering the same thing.  The answer I keep coming away with from the "heavy-hitters" here seems to me to be one of control over the process and results.  BPP does indeed produce an "acceptable" integrated image, but manual image integration allows you to play with the results a bit more.  But as you have pointed out, there seems to be enough controls in BPP that you should be able to get close to the same results.

I know this wasn't really an answer, but maybe it will shine a brighter light on the issue to those who can make more sense out of it for us.

[RickS]

It's a matter of convenience.  If you want to just do the Integration several times while adjusting the rejection parameters it is easier and quicker to run ImageIntegration several times instead of the whole BPP script.

Cheers,
Rick.

[jerryyyyy]

My impression is that the image integration controls are the same on the script, but if you run the script you have to go through the whole nine yards, i.e. calibration and registration too.... [read lotsa time]  I suspect you can run the script to get the registered files and then mess with the integration settings just running the last step, i.e. integration... [read quicker]  I have never gotten much mileage from this, but probably am missing some Insight in my Pix... get it?

[jkmorse]

Be careful about using the batch processing tool for Image Integration.  In fact Juan says never do it for production work.  In that regard, you might find the following discussion helpful:

http://pixinsight.com/forum/index.php?topic=6309.msg42975#msg42975

Best of luck and stick with it.

Jim

[Warhen]

BPP is pretty damn powerful, and as Cdeselles mentions, results look quite good. I think the most obvious limitation of the script is that you can't choose more than one rejection algorithm, So, if for instance, you have many L frames where Linear Fit clipping would be appropriate, but you only have 9 frames of each color channel, where Average would be a better choice, you have to compromise on one.

You also can't use more than one master dark or bias at a time, and though the script is capable of matching different temp lights with matching reduction frames, it can only use one master bias or dark at a time. And, if these limitations impact your particular data, going the ImageIntegration route, does indeed makes more sense.

[cdesselles]

BPP is pretty damn powerful, and as Cdeselles mentions, results look quite good. I think the most obvious limitation of the script is that you can't choose more than one rejection algorithm, So, if for instance, you have many L frames where Linear Fit clipping would be appropriate, but you only have 9 frames of each color channel, where Average would be a better choice, you have to compromise on one.

You also can't use more than one master dark or bias at a time, and though the script is capable of matching different temp lights with matching reduction frames, it can only use one master bias or dark at a time. And, if these limitations impact your particular data, going the ImageIntegration route, does indeed makes more sense.

So Warren, would it be appropriate to run each channel through BPP individually along with thier respective calibration frames resulting in 4 seperate Master Lights, then use Image integration to combine the 4 images into the final result for further processing?

[Warhen]

I'd say 'appropriate' yes, but optimal- no. And this is because Juan will recommend using ImageIntegration to stack, and frankly, at this point considering the extra steps involved, why not? That being said, my opinion would be sure, do multiple passes with the script if you're comfortable there. You can then take advantage of specified rejection algorithms. Now, just to clarify your question regarding 'each channel'- the script can handle multiple channels, it's just limited to one rejection algorithm per pass.

[jerryyyyy]

I have come to a realization of where this might be useful, aside from adding the drizzle files. 

I have some DSLR (Nikon D800) files with a lot of noise.  I can calibrate and register the files in BPP, but then use ImageIntegration and change the parameters of the rejection algorithms to see what works best.  This I do on the calibrated and registered files produced by BPP> 

[bhwolf]

Not sure if anyone else does this, but lately I've switched BPP to do no rejection for stacking the registered lights.  I've found it helpful to use this image as a baseline ... I think this would always have the highest SNR but likely unusable final product.  I then use Image Integration on the registered images with rejection to gauge the results to make sure outliers are effectively rejected without rejecting diserable signal. 

When starting from a likely good but not optimized master light generated by BPP (using, say, WSC) I felt like I was working backwards, dialing back rejection until I started seeing outliers.  Since this is a judgement call, having the no-rejection average to start with works best for me. 

It is very quick to restack the registered frames BPP produces ... Doesn't really make sense to go through BPP again unless you are recalibrating. 

It is kind of fun, too, to see how much the right rejection routine can improve the image!

[jerryyyyy]

Not sure if anyone else does this, but lately I've switched BPP to do no rejection for stacking the registered lights.  I've found it helpful to use this image as a baseline ... I think this would always have the highest SNR but likely unusable final product.  I then use Image Integration on the registered images with rejection to gauge the results to make sure outliers are effectively rejected without rejecting diserable signal. 

When starting from a likely good but not optimized master light generated by BPP (using, say, WSC) I felt like I was working backwards, dialing back rejection until I started seeing outliers.  Since this is a judgement call, having the no-rejection average to start with works best for me. 

It is very quick to restack the registered frames BPP produces ... Doesn't really make sense to go through BPP again unless you are recalibrating. 

It is kind of fun, too, to see how much the right rejection routine can improve the image!

This sounds very reasonable.  This AM I am working on an image where I got out RGBL images, then got the noise for the L from the BPP default Linear integration 7/5 parameters on the cuts.  I then integrated with the drizzle files using ImageIntegrator and included all the LRGBs (to make Super_L).  Bottom line is noise was an order of magnitude less in the Super L. 

I think you could do what you suggest and then just manipulate the parameters and check what comes out for noise as an objective measure of quality.  Probably an overly simplistic idea....

[RickS]

Not sure if anyone else does this, but lately I've switched BPP to do no rejection for stacking the registered lights.  I've found it helpful to use this image as a baseline ... I think this would always have the highest SNR but likely unusable final product

This process of careful integration is a well known technique that I have been using since before BPP existed.  You can use the SNR estimate from an ImageIntegration without rejection as a target.  I first read about this in the presentation "Image integration techniques: Increasing SNR and outlier rejection with PixInsight" Presentation at "VI Seminario de Astrofotografía de Cielo Profundo"  Madrid, 20th November 2010" at http://www.astrosurf.com/jordigallego/articles.html

Cheers,
Rick.

[vpcirc]

In reviewing older tutorials, Harry mentions selecting your reference image in Image Integration. When you do, it's moved to the stop of the stack.  I'm not an advanced user by any means, but my instinct tells me that may give the reference for other images in the stack to compare too. This seems very similar to how CCDStack weights every image based on the signal in the reference image. If this is the case with the math, then BPP would be a poor choice for making a final image unless every frame is pretty darn close to the same. I've always wondered why a pattern seen in a few frames shows up in the final frame as my thought is that data should be rejected. If the frame on top happens to contain it, why would it if the algorithm starts out as seeing as signal that should be there? I am not saying this is fact or proper procedure, I'm experimenting and learning. I would welcome Juan's input.   

[jerryyyyy]

Are you all getting the noise estimates from the Image Analysis Script NoiseEvaluation? 

[RickS]

Are you all getting the noise estimates from the Image Analysis Script NoiseEvaluation?

I use the "Median noise reduction" value that ImageIntegration writes to the Process Console.