“What is a mosaic image?” people typically ask when I use the term. I explained it before, so I’ll just c&p:
Mosaic images
A mosaic is created by stitching (planar) photos together where each picture is taken from a different location, with the camera pointing in different directions. A special, easier case is stitching flat scans (or images where the camera moves in relation to the object, but has the same distance and points in the same direction). This method can be used to remove objects in the foreground, by applying masks or cutting out parts of the pictures. If distortion is removed from each image singly before they are combined, this is also wonderful for creating high-res images of objects, or for getting totals of objects you can’t step far enough away from to take one photo (AMNH Styracosaurus comes to mind).
But what about skeletal mounts? That’s what us dinosaur palaeontologists are into, after all. For them, the mosaic technique is not that well suited, because objects at different distances will experience parallax shifts between images. And this makes it very hard to create a good complete view, because you’d have to split each image into several, based on how far what part of it is from the camera. For aerial surveys of quarries, however…. a nice juicy bonebed….
This already gives you a number of hints about the requirements, including:
- photos should be taken at equal distance from object
- photos should be taken with the camera positions absolutely parallel to each other
- photos should overlap a lot
If you do not follow these rules you induce a lot of parallax issues or can’t deal with the inevitable ones, and that can ruin your final result.
Here’s how I usually try to do it:
Taking the photos
Ideally, I have help for taking the photos. A tripod is nice, and a second person who helps me place it correctly. Or I simply use architectural guides, such as the lines between tiles on the ground, or railings, or edges of steps, or whatever I can find. The aim is to have the camera always at the same distance from the object, and any help I can get I use. Don’t be too shy to ask two complete strangers to spend five minutes holding a piece of string that serves as a guideline for similar distance. Don’t be afraid of looking totally silly when you take measured steps (I usually do one shoulder-width sidestep for close objects, two to three of them for stuff far away), or measuring distances out by placing your feet heel to toe. It is worth the effort and ridicule! Besides, if you snap pictures with an important expression on your face people will think you’re a professional photographer, not a moron
When you can get it, try to have the full height of you object in each photo. If not, either do lateral rows, or vertical ones – but remember, for a true mosaic flat scan you need to do the rows not by pointing the camera down and up, as I did for the monster crinoid shown again below (which explains the slight errors you can see on the upper and lower 25% of the stitch), but actually have the camera at two or more different heights, and point the lens at the object at a perfect right angle. Thus, on your knees and standing up, or using a ladder…… it can take a lot of effort! I find that usually, two well-overlapping lateral rows (i.e., each covers >65% of the other) can tolerably be done by angling the camera up and down, if the object is really flat. Less overlap for taller objects requires three rows, and you quickly get to the point where it is worth doing four and a true flat scan shoot.
Preparing the stitch
Once you have your photos done you can go straight into Hugin. Load all the photos (and there will be a lot more than you need for a panorama), and see what the assistant does. Just follow this tutorial. If you are lucky, if your hugin is configured correctly, and if your PC has enough oomph, that’s it! Do the usual playing around with the suggested panorama, and stitch! I usually stitch once at a very low resolution, e.g. 1/5 of full res, to see if there are major screw-ups. If so, do as below. Otherwise just run it at full size and enjoy!
Troubleshooting
One of the biggest annoyances with this topic is that hugin will often not be able to find control points perfectly between >30 images. On my Laptop it normally balks at >8 photos . Thus, I do the fix described in this post: I select groups of photos that I know overlap and manually start the control point finder for them.
One thing that is really helpful is the Fast Preview Panorama window. Once you have created control points and aligned the images you can use it to check if there are overlapping images with no control points – the main source for problems with the final stitch. If there are jumps in the image, such overlapping but non-controlpointed images are the reason! Go to the ‘Layout’ tab of the Fast Preview Panorama, use the “scale” slider to make the image thumbnails really small, and lo and behold! lines appear between them that show you the status of the alignment: overlapping images are connected by lines, and if they are thin and white or grey, no control points exist between these two images (red, yellow and green lines means few, some, and plenty of CPs exist).
You can now simply click such a line and hugin will bring up the two images in the ‘Control points’ tab. Remember their numbers (at the top), go to the ‘Images’ tab, select them, and use the auto-CP-finder to make hugin find CPs for only this pair. Usually, this works.
If not, well, tough luck: you will have to manually place CPs in the ‘Control points’ tab (explained here, end of post). Takes a few minutes, but is totally worth it.
Once all image pairs are connected by red lines at least you can try again to optimize and stitch. Should be better now. However, it is entirely possible that the CPs cluster at one end of the photo, and jump lines appear at the other end of the photo in the final image. Two solutions exist: manually add CPs at the “empty” end, or use a mask to crop away one of the two photos in that area. I usually try to former.
(in case you wonder: the above example with all the unconnected images – I didn’t do anything with it! It stitched perfectly despite all the missing CPs! Don’t believe me? see here…..)
As a small break from all the words here is again the first mosaic I made where I used the Fast Preview Layout tab. And where I needed it, too!
OK, all that done, hat’s next?
Masking
Very often there are railings, lamp posts or other things in your view, and obviously not being on the same plane as the object you photograph they will suffer from parallax issues. The solution is simple: get rid of them, or at least get rid of all instances but one. Ideally, if you want to retain them you should retain the version that is as close to the center of a photo as possible. Use the ‘mask’ tab of hugin to draw masks closely around the annoying objects, and you’re done.
Now stitch and enjoy!
Oh, for all those who read through the full post I have a special extra hint on taking good photos perfectly parallel: if your intended object is behind glass, take the lens hood off your camera and put it flat against the glass. Perfect orientation and distance right there
That’s what I did for all the behind-glass stuff at the Urweltmuseum Hauff, and it worked very well.



Thanks for this, Heinrich.
I love hugin but it suffers from lack of documentation and tutorials.
And since it is so feature-rich, that is particularly unfortunate.
Your series of posts filled a huge gap and I thank you for the well-done efforts.
you are very welcome!
In fact there are quite a lot of tutorials etc. out there, but mainly for the componetns that hugin is the user interface for. Thus, for a computer noob they are hard to find and understand.