Lens correction and simulating larger sensor


#1

I’ve been thinking about stitching nine to 12 images to simulate a larger sensor. I have a Canon 400D with a kit lens 18-55mm. Assume I should shoot at 55mm to minimize distortion. I have successfully done some stitching (Hugin) with files from the smart phone camera, but they were all across (or in one instance vertical). Is there anything I need to know about putting together multiple exposures like this? I have a sturdy tripod with built-in levels.


(Ingo Weyrich) #2

The 400D has a sensor size of about 3888x2592 pixels for one shot.

Rotating the cam by 90 degrees and calculating with a 20% overlap in horizontal direction that gives about 6740 x 3888 pixels. Now lets’ cut 10 % from the 3888 because you didn’t level correctly. Still 6740 * 3500 pixels with 3 shots.

Easy rule: If you have for example a 35mm lens on your cam and want to make a stitch which reflects a 24mm lens, just turn the cam by 90 degress, make three shots with a bit overlap and stitch them together. As a benefit you will also get increased resolution compared to one shot whith (in this example) 24 mm lens.


#3

You don’t need to shoot at a longer focal length to minimize distortion. If you overlap the shots enough, Hugin can calculate the distortion of the lens and correct for it.

I prefer to overlap 50% or so.


#4

CarVac that may be true with DSLRs. But I had some smartphone images that caused Hugin to choke. Presumably too much curvature?


#5

No smartphone I’ve used has very much distortion…

You have to use the advanced mode and choose the appropriate fit parameters.


(Mica) #6

I’ve had this thought as well, mostly just because I want to make a really huge image. For this reason I put together a panorama tripod head so that I hopefully would have an easier time stitching. So far I have not had success; the power lines in my single row, 360 degree panorama never line up.


#7

I will look into the parameters CarVac. But too bad Hugin won’t read my mind

Paperdigits the original training I got in photography said lens distortion is at its worst near the edges of the field of view. So I’m thinking I should crop carefully before I put everything into Hugin? Just as long as some overlap remains.

Question:
When I do assemble, should I (for example) stitch the horizontal rows, and then do another stitch for the vertical assembly?


(Isaac Ullah) #8

The technique is called “multi row panorama”. This is a great article on the topic that you may find helpful: https://focusphotographers.org/shooting-multi-row-panoramas/

Hugin usually does this very well on “auto” mode. I find that putting a “horizontal line” type control point along the horizon to be very helpful in ensuring a good stitch for this type of multi row pano.


#9

Oh wow that is really good! Thanks for posting.


(Morgan Hardwood) #10

You need a pano head, otherwise you will have parallax issues.

As stated by others, lens distortion is very unlikely to be the problem. Not rotating around the nodal point is the most likely cause. Ergo pano head.

Forget about lens distortion, and don’t crop your images. Lens distortion is not a problem when dealing with programs like Hugin.

You need around 20% overlap.

If you’re dealing with a problem, don’t leave us guessing - upload the image.

No. Load all at once so that you detect control points across all at once. That will improve the overall stitch quality and help Hugin better figure out the lens distortion parameters.


(Pat David) #11

I’m assuming you are wanting to do something similar to the so-called Brenizer Method?

If so, to help maximize the effect it might be good to shoot at a wide aperture (or however you’d like to maximize your DoF). Shooting on the long-end can help minimize parallax problems (assuming objects of interest and in focus are on a similar plane - this is not guaranteed and will not fix parallax, just hopefully minimize it).

I’d defer to the already great answers from everyone else. If you’re shooting a subject, I can recommend shooting your first frame(s) on them in order to freeze any motion. From there I’ve generally shot in a circular path from there around them (not much regard for a concept of “rows” or “columns” :smiley: ).

I can point you to a neat tool that will let you calculate your equivalency for a larger sensor (I came across this a while ago and have not vetted its accuracy):

http://brettmaxwellphoto.com/Brenizer-Method-Calculation/


#12

OK I’m in the midst of the learning curve, thanks to all the pointers! There is at least one site dedicated to pano photography, and also this related wiki:
wiki at panotools dot org


(Isaac Ullah) #13

If you are only shooting far away subjects (like a landscape), then parallax will not be as big of an issue, and you can get away with a simple rotary tripod head. A “nodal slider” is better, of course, and a true pano head is the best. If you are going to include any foreground elements, then you really need one. The good news is that it’s easy to jerry-built one with a cheap macro slider rail and a L flash bracket. You can get that setup for about $20 from Amazon…


#14

Isaac that’s a nice summary. Of course I would buy top of the line if I could, but “cheap” is an acceptable entry point for a guy like me, experimenting. Thanks!


(Isaac Ullah) #15

Since a picture is worth a thousand words, here’s exactly what I’m talking about:

L flash bracket and macro rail slider. I also added a cheap “arca Swiss” type quick release plate on the bottom of the L bracket, which gives me a little bit of left to right play too, to get the lens lined up perfectly with the rotation point of my tripod head. All of these things are the cheap “Neewer” brand that you see all over Amazon and eBay these days. They aren’t as good as the more expensive brands, but perfectly serviceable, and waaaay cheaper.


(Mica) #16

I’ve out something similar together for just under $250. Still trying to find the no paralax point :stuck_out_tongue:


#17

If the foreground element is not part of some overlap, or at least entirely on one of the frames and leaves enough overlapping structures to be masked out of adjacent images, it could* work without.

*Disclaimer: Just theory, not tested.


(Isaac Ullah) #18

Yes, you are probably right. I think you’d have to use a mask to ensure it showed up fully, but if it’s all in one frame, then it should be ok…