Mosaic Processing

[Mav359]

Just Curious

When you do a mosaic of an object do you align, stack, calibration subtraction & histogram stretch each Panel & then stich them together

or do you just keep gathering data, stich them first and then do the rest?

cheers

[martin_h]

I calibrate and stack them then stitch them, then process as a whole image.

[sologuitarist61]

That is interesting. I did a simple two pane of the sun the other day (top and bottom halves) and they were taken immediately after each other on the same settings, yet both had different tonal values etc. They stitched together OK but the difference between them was noticeable evn before expanding the image size.

[ollypenrice]

I get two (or more) stacked images together, stretch them in parallel, while trying to keep them close together in Levels and Curves, then I place one over the other and get them as close as I can before stitching.

Olly.

[Mav359]

Cheers guys

Deffo going to have to learn the process of mosaics as my setup has quite a small FOV

[sologuitarist61]

I get two (or more) stacked images together, stretch them in parallel, while trying to keep them close together in Levels and Curves, then I place one over the other and get them as close as I can before stitching.

Olly.

Sorry Olly being a bit thick here - how do you stretch them in parallel? I think that was my problem, that the stretching seemed to affect both halves differently.

[IanL]

What you are

Sorry Olly being a bit thick here - how do you stretch them in parallel? I think that was my problem, that the stretching seemed to affect both halves differently.

Some part of your first image will overlap part of the second image (if they don't then you need to change your mosaic technique so that there is a reasonable overlap between panels).

Your objective is to get each element of the overlapping parts to the same brightness level; this is easier with mono images as you only have to match the corresponding pixel brightness values between the two images. It can be slightly more complicated if you are trying to match two colour images if they have differences in hue, since you then need to match the R, G and B values for corresponding pixels.

Start simply with the black point and white points and try to match those first. So you would open up both images side by side, and bring up the histogram/curves tool for one of them. Find a background area that appears in both images and use the appropriate tool to measure the pixel values in the first image (you can just sample a few points to look at the pixel brightness, or if you have a package that can do stats, select a small area and calculate a mean or median value). Now go to the curves tool for the second image and set the black point of the curve so that you end up with the same brightness value for the corresponding area in the image.

Now do the same thing for the brightest part of the image. Measure the brightest area in the first image where there is an overlap, adjust the white point in the second image to get the corresponding pixels to match. Note that this is much easier to do if you have the brightest part of the image in the overlapping area. If the brightest part of the image is not common to both, then I would use the one with the brightest part as my first (sample) image, and adjust the second (darker) image. If you do it the other way round you may find you start blowing out the highlights without realising it until later.

Once you have established suitable black and white point values, make a note of them so that you can quickly start over if you need to. Now you just need to add more points to the curve to get the mid-tones to match in the overlapping areas. Its the same routine, sample the first image, adjust the second image to get the corresponding area to match. The trick is to try to keep the curve really simple if you can; once you have matched the pair of images as best you can, my approach would be to bake those changes in to the second image, composite the pair of images and apply any further curve adjustments to the assembled mosaic.

Assuming you got the first stage of matching right, you should see that any further curves adjustments apply evenly across the whole image. if you find that the seams between the two panels start to show up (usually obvious in the background areas), then go back to the initial matching stage and try again.

This isn't the only way to do it, and you could apply all of the curves adjustments to both images prior to compositing them if you want, but by breaking it down in to stages you can understand what you are trying to achieve more easily and spot any problems earlier in the process.

[sologuitarist61]

Thanks Ian, I have copied that off to go in my Documents folder so that I can refer back to it when required. Many thanks

[ollypenrice]

I agree with Ian's method though if you just set the black and white points and apply similar curves to each image they should fit together fairly well. One thing, though; I would always get them fully stretched before combining because, if you don't, further stretches may well make any joins visible again. Where gradients cause images not to fit neatly you can use the dodge and burn tools judiciously to make them behave!!

Olly

[sologuitarist61]

Good point, thank you Olly, your comment will be added to the aforementioned word document!

[milosz]

Mav,

With PixInsight you first stich the unstretched panels and than do the processing of the assembled image. Here is the tutorial:

http://pixinsight.com/videos/StarAlignment/Mosaic/en.html

You can also do the mosaic using gradient tools from PI. They work much better than the tutorial above. Here is the tutorial about that:

http://www.harrysastroshed.com/Mosaic.html

milosz