What means the statistics ?

[Dominik]

Hello !

I'm using Pixinsight since a few months, and I'm trying to  understand what mean the statistics after using ImageIntegration, for example as on this image : 

.

If I want to maximize the signal to noise ratio, what numbers must i examine in this image ?

Thank you !

[Alejandro Tombolini]

Hi Dominik, welcome to the forum

Please, refer to "Console Statistics" in the documentation of the Image Integration tool. I think that you will find the answer there.

Saludos, Alejandro.

[Dominik]

Thank you for you answer !

First, i have 3 values so I suppose (because I have images coming from my 350D) that each value concerns R, G and B channels ?
 
Well, i found that in the doc :
The reference noise reduction value is relative to the reference image (the first input image), while the median noise reduction is the median of computed values for all images; this is the most significant value that should be minimized to achieve the best possible SNR increment.

So I understand that i must try to obtain the smallers values as possible in the line median noise reduction ?

To obtain these statistics, I made a ImageIntegration with no rejection, thinking that i will obtain the best SNR as possible. Am I right ?

Then, i made a second ImageIntegration (with my 30 images) using a Linear Fit Clipping, and I obtain this result :



I see that the first value (2,228) is now higher (so the snr of the R channel is lower ?), and the second ans third values are lower (so the snr of the G and B channels are better ?) ...
Well, it's not clear for me :
- the R value can increase while the G and B values can decrease ?
- I thought that a Integration whith no rejection would always produce a better snr (so lower value on all channels ?) that a integration with rejection, right or wrong ?
- what's the way to improve the snr on all channels ? Or must I try to improve the snr on one particular channel depending of my image ?

Sorry for all these questions ...

[Juan Conejero]

Hi,

It's just the opposite: noise reduction has to be maximized. This is actually an error in the latest version of this document. It should read:

this is the most significant value that should be minimized maximized to achieve the best possible SNR increment.

Sorry for the trouble. I'll fix this error as soon as possible.

[fulatoro]

Juan,

I am glad I just found this post, I was going to ask about some confusion in the documentation as I was also trying to use the Median Noise Reduction to compare different pixel rejection algorithms on my lights. I was going to mention that in the document, we see the two statements below:

this is the most significant value that should be minimized to achieve the best possible SNR increment.

Further below we read:

Your goal when integrating a set of light frames is to achieve the maximum possible noise reduction with the necessary rejection of outlier pixels.


I guess there is indeed an error in the documentation.

I have been enjoying learning PixInsight immensely.

[Geoff]

I have been enjoying learning PixInsight immensely.

I had a similar experience
http://pixinsight.com/forum/index.php?topic=4247.msg29799#msg29799
I don't know why people keep complaining about the "steep learning curve"; it's fun to explore.
Geoff

[pfile]

based on what i see on CN, a lot of people are not interested in understanding the 'why' of what they are doing and instead just want to know what series of things to click on in order to get a result. it really does not matter what software package you are using - that's a recipe for lifelong ignorance

PI does not lend itself well to just clicking on stuff, you need to have an understanding of what you are doing and why…

rob

[fulatoro]

I agree totally.  It takes a bit of time to understand the theory behind it, but I believe doing so helps you better understand your images and what tools might help address certain issues.

While I was testing out the different Pixel Rejection mechanisms, I ran into an interesting situation...I tested Linear Fit, Averaged Sigma clipping and Winsorized Sigma Clipping to try to determine which was doing a better job at integrating a set of 10 M33 Lights.

Looking at the pixel rejection percentages, I saw that WSC was rejecting a lot more pixels than the other two algorithms. Looking closely at the images, it turns out that there was a 6 pixel black border around my lights.  This was only rejected by WSC. I had not paid attention to it, but I suppose I could have DynamicCrop them later even with the other rejection mechanisms failing at excluding.  It was however a tangible proof that the WSC rejection mechanism did perform better then the other two and ultimately it resulted in a higher Median Noise Rejection as well.

The question is this...Is this a false positive biased towards WSC because of its ability to reject the black border such that I get a higher Median Noise Rejection but not necessarily a less noisy image. My understanding based on the literature (See this awesome Powerpoint from  Jordi gallego at http://www.astrosurf.com/jordigallego/articles.html 

"Image integration techniques: Increasing SNR and outlier rejection with PixInsight" , an excellent read by the way, dates from 2010 but still very relevant, he states that for lights numbering between 5-10, Average Sigma clipping is recommending, 10-15 WSC, > 15 Linear fit.


checkout the Noise Reduction talk as well while you are at it.


Moussa