Image pixel size for displaying on the forum

[Chefgage]

Hi everone. What is a good size pixel wise for uploading/displaying in the forum. My images are 6000x4000 but to display them without having to scroll across would be better. Or does the device that is displaying them sort of resize the image to fit the screen you are looking at be it a mobile or large tablet etc. I did used to know all this but its been a while since i really uploaded any images to forums.

Thanks.

[happy-kat]

This website cleverly makes the image fit the browser size. Jpg or Png are nice for web viewing.

Don't you bin your images during processing? Or are you talking about a mosaic.

I generally bin 50% so the final size both physically and file size is nice and fairly small.

Edited February 19, 2020 by happy-kat

[vlaiv]

Hi, you need not worry about display size - it will always be "fit the screen" by default, regardless of resolution of underlying image.

Some people, me included, like to view image at 1:1 setting or 100% zoom, even if I need to pan around and that is possible if you open image by itself in browser - I usually hit right mouse button and do - open in new tab ...

At first image will be scaled to fit the window again - but simple click on it will expand it to full size and you can pan around.

I mention above because of what I'm about to say next. There is proper resolution for astronomy image, or rather range of proper resolutions. Sometimes with modern cameras and larger telescopes (longer focal length) people make image that is just too zoomed in when viewed at 100%. Stars are no longer small dots but rather "balls" suspended in space and everything starts to look blurry on 100% zoom viewing. It will still be nice image to look when it is viewed scaled to screen size.

Such images are worth down sampling to appropriate size as it will make viewing at 100% more enjoyable. Luckily there is simple technique to determine proper sampling rate for the image - you need to measure star FWHM (Deep sky stacker gives you this information for each frame) and divide that with 1.6.

If your star FWHM is 4px then 4/1.6 = 2.5 - you need to resize your image to 2.5 smaller size. If you get number that is less than one - just leave image as is - don't enlarge it as it will be blurry (enlarging won't bring missing detail back in).

[happy-kat]

Is the x1.6 significant to camera sensor size or a universal figure to use for any camera?

[Chefgage]

23 minutes ago, happy-kat said:

This website cleverly makes the image fit the browser size. Jpg or Png are nice for web viewing.

Don't you bin your images during processing? Or are you talking about a mosaic.

I generally bin 50% so the final size both physically and file size is nice and fairly small.

If by bin you mean cropping then no i do not. But i have only just started doing astro photograpghy really so still learning the basics.

[Chefgage]

24 minutes ago, vlaiv said:

Hi, you need not worry about display size - it will always be "fit the screen" by default, regardless of resolution of underlying image.

Some people, me included, like to view image at 1:1 setting or 100% zoom, even if I need to pan around and that is possible if you open image by itself in browser - I usually hit right mouse button and do - open in new tab ...

At first image will be scaled to fit the window again - but simple click on it will expand it to full size and you can pan around.

I mention above because of what I'm about to say next. There is proper resolution for astronomy image, or rather range of proper resolutions. Sometimes with modern cameras and larger telescopes (longer focal length) people make image that is just too zoomed in when viewed at 100%. Stars are no longer small dots but rather "balls" suspended in space and everything starts to look blurry on 100% zoom viewing. It will still be nice image to look when it is viewed scaled to screen size.

Such images are worth down sampling to appropriate size as it will make viewing at 100% more enjoyable. Luckily there is simple technique to determine proper sampling rate for the image - you need to measure star FWHM (Deep sky stacker gives you this information for each frame) and divide that with 1.6.

If your star FWHM is 4px then 4/1.6 = 2.5 - you need to resize your image to 2.5 smaller size. If you get number that is less than one - just leave image as is - don't enlarge it as it will be blurry (enlarging won't bring missing detail back in).

Thanks for that. I will have a look on deep sky stacker to see what it says.

[vlaiv]

15 minutes ago, happy-kat said:

Is the x1.6 significant to camera sensor size or a universal figure to use for any camera?

That 1.6 is related to Nyquist sampling theorem and is approximately sampling rate that corresponds to given FWHM. If FWHM is expressed in arc seconds, then dividing with x1.6 will give you sampling resolution in arc seconds per pixel. If FWHM is expressed in pixels - then dividing with 1.6 will give you "pixel size" - or factor to reduce your image by. In above example if FWHM is 4px then ideal pixel size is 2.5 - so you need to reduce image by factor of x2.5

5 minutes ago, Chefgage said:

If by bin you mean cropping then no i do not. But i have only just started doing astro photograpghy really so still learning the basics.

Binning is form of resizing the image down and has some advantages over regular resizing down. It reduces noise better. Every resizing down of the image reduces the noise in the image, but different resizing methods reduce noise by different amount. Binning just adds adjacent pixels together and forms one large pixel out of group of 2x2 or 3x3 pixels. For this reason it can only resize down by integer factor - either by x2 or by x3 or x4, etc ... (it can't reduce size by x2.3 for example). However such reducing down has very good effect on the noise - it also reduces noise by x2, x3, x4 .... etc and it is mathematically predictable in the way it changes noise (no correlation between pixels, always exact improvement in noise and such) and it is therefore preferred way to do things in astronomy (science side of things).

[Chefgage]

2 hours ago, vlaiv said:

That 1.6 is related to Nyquist sampling theorem and is approximately sampling rate that corresponds to given FWHM. If FWHM is expressed in arc seconds, then dividing with x1.6 will give you sampling resolution in arc seconds per pixel. If FWHM is expressed in pixels - then dividing with 1.6 will give you "pixel size" - or factor to reduce your image by. In above example if FWHM is 4px then ideal pixel size is 2.5 - so you need to reduce image by factor of x2.5

Binning is form of resizing the image down and has some advantages over regular resizing down. It reduces noise better. Every resizing down of the image reduces the noise in the image, but different resizing methods reduce noise by different amount. Binning just adds adjacent pixels together and forms one large pixel out of group of 2x2 or 3x3 pixels. For this reason it can only resize down by integer factor - either by x2 or by x3 or x4, etc ... (it can't reduce size by x2.3 for example). However such reducing down has very good effect on the noise - it also reduces noise by x2, x3, x4 .... etc and it is mathematically predictable in the way it changes noise (no correlation between pixels, always exact improvement in noise and such) and it is therefore preferred way to do things in astronomy (science side of things).

From a GIMP point of view is this what it refers to as interpolation. Using the linear, cubic method of resizing.

[vlaiv]

2 minutes ago, Chefgage said:

From a GIMP point of view is this what it refers to as interpolation. Using the linear, cubic method of resizing.

Indeed - when I say there are different methods of resizing - I mean interpolation. Binning is sort of interpolation algorithm, but it works only on integer down sampling factor.

In fact - binning 2x2 is mathematically equivalent to 50% size reduction + (0.5px, 0.5px) translation with use of bilinear interpolation method (average of 4 adjacent pixels is the same as linearly interpolating at point that is corner joining all 4 pixels - if pixels were considered to be squares).

I mentioned one advantage of binning - no correlation introduced. Don't know how much you know about interpolation, but different interpolation algorithms use more than just surrounding pixels to calculate value when resampling - bilinear uses 4 nearest pixels, but bicubic uses 16 pixels around wanted point (matrix of 4x4 - as you need 4 values to fix cubic function, 3 values to fix quadratic function and only 2 to fix linear function). There are of course other interpolation methods and each of them will have different properties with respect to details in the image and impact on noise.

[Chefgage]

1 hour ago, vlaiv said:

Indeed - when I say there are different methods of resizing - I mean interpolation. Binning is sort of interpolation algorithm, but it works only on integer down sampling factor.

In fact - binning 2x2 is mathematically equivalent to 50% size reduction + (0.5px, 0.5px) translation with use of bilinear interpolation method (average of 4 adjacent pixels is the same as linearly interpolating at point that is corner joining all 4 pixels - if pixels were considered to be squares).

I mentioned one advantage of binning - no correlation introduced. Don't know how much you know about interpolation, but different interpolation algorithms use more than just surrounding pixels to calculate value when resampling - bilinear uses 4 nearest pixels, but bicubic uses 16 pixels around wanted point (matrix of 4x4 - as you need 4 values to fix cubic function, 3 values to fix quadratic function and only 2 to fix linear function). There are of course other interpolation methods and each of them will have different properties with respect to details in the image and impact on noise.

Again, thank you very much for your detailed reply. From how much do i know about interpolation - very little. But as with most things i am interested to learn 😀

Just need some clear (and not windy) skies to get some data and put it into practice.

Edited February 19, 2020 by Chefgage

[ollypenrice]

Quite apart from sampling theory, there is the matter of exposure time to consider. Even assuming you are not over-sampling (by shooting at less than about an arcsecond per pixel) you will need a lot more integration time to produce an image which looks good at 100% than at a reduced size. As Vlaiv said, if you are over-sampling the image will never hold up to scrutiny at full size (when one camera pixel is given one screen pixel.)  But even if you are under-sampling it will take you longer to get a signal to noise ratio that will support full size scrutiny than one to be viewed at reduced size.

In general I try to reach an S/N level which will support full size and sometimes, when imaging small targets, it's essential to do so to get the target to a respectable screen size. However, in shooting a widefield mosaic, for instance, you might feel that there was no need to make the final image of several panels hold up to full size. You might prefer to reduce the exposure per panel, resample the panels downwards and still end up with a very large picture.  This might take the panels down from 10 hours exposure to 5 with negligible loss once shown at considerably reduced size.

It's part of the planning, therefore, to decide in advance what your target presentation size will be. Of course, if you want to play it safe you should follow the advice of SGL member Maurice Toet who, when asked how long he intends to expose for, replies, 'As long as it takes!'

Olly

[wornish]

This is a very interesting thread can I check my understanding ?

I have an image that is 4616 x 3488 pixels and the FWHM is on average about 3.5

So if I use the formula for best image size then I should resize the image by 3.5/1.6 = 2.185

ending up after resizing at  2110 x 1595px

Is this correct ?

[vlaiv]

16 minutes ago, wornish said:

This is a very interesting thread can I check my understanding ?

I have an image that is 4616 x 3488 pixels and the FWHM is on average about 3.5

So if I use the formula for best image size then I should resize the image by 3.5/1.6 = 2.185

ending up after resizing at  2110 x 1595px

Is this correct ?

Provided that FWHM is reported in pixels - yes.

In this situation, I would actually bin x2 and then check FWHM and maybe scale by a bit, but I don't think that you would need to do any scaling as difference would be minimal. Do bin while data is still linear prior to processing.

If you have 3.5" FWHM - then you should aim at about 2.185"/px sampling rate.

[Xsubmariner]

I have enjoyed reading this thread, thanks Vlaiv for your straight forward words of wisdom. Now the confession, as a relatively new imager I have started capturing your words of wisdom in a notes document “ Vlaiv’s words of Imaging Wisdom” as a reference to remind myself in these early stages. Hope you don’t mind.

Martin

[vlaiv]

10 minutes ago, Xsubmariner said:

I have enjoyed reading this thread, thanks Vlaiv for your straight forward words of wisdom. Now the confession, as a relatively new imager I have started capturing your words of wisdom in a notes document “ Vlaiv’s words of Imaging Wisdom” as a reference to remind myself in these early stages. Hope you don’t mind.

Martin

Of course I don't mind. Wanted to do something similar myself for quite some time now, but never seem to find spare time to do it. I wanted to make a website / blog kind of thing and do my best to explain and demonstrate things related to astronomy (mostly imaging/processing/theory). Hopefully I'll get it started sometime soon.

[wornish]

1 hour ago, vlaiv said:

Of course I don't mind. Wanted to do something similar myself for quite some time now, but never seem to find spare time to do it. I wanted to make a website / blog kind of thing and do my best to explain and demonstrate things related to astronomy (mostly imaging/processing/theory). Hopefully I'll get it started sometime soon.

Excellent idea.  You would have lots of readers.