Photography - Stitched images

February 28, 2013  •  Leave a Comment

Why do we need to stitch images?
There are are two reasons. 1. For panoramas and 2. For higher resolution, though it can be a combination of the two.

As digital cameras get more sophisticated and higher in resolution, it might be thought that stitching images isn't really necessary any more. Several cameras and smart phones have panorama functions built in - and there are numerous apps, some of which which work very well. Image resolutions seem to go up inexorably, but does all that add up to better images? Being a photographer, and though I love my iPhone and use it to take pictures and now panoramas quite often, I am always aware of its limitations.

Smart Phones and Small sensor cameras
There are a few weaknesses with built in panorama functions of these smaller devices, including dynamic range issues, where highlights can be blown out for example, and movement which has to be managed by appreciating how the devise works - rather than a combination of planning the shots and then having control at the stiching stage as one does with single images stitched afterwards.

The iPhone 5 for example scans a vertical strip of 'view' as the camera is rotated to capture the image. This works quite well, as it's less susceptible to subject movement - though it isn't faultless. Some cameras have more problems with moving subjects and it should be said that no panoramic device really works well with moving subjects - unless it is a highly specialised device.

This image shot with the iPhone 5 shows how its' JPEG output can't capture the highlights in the valley fog properly, and it's tiny lens and aperture creates a rather ugly sun. The rest of the image is pretty good and is 28 megapixels in size! 

Madonna Skiing 10840

More control..
Micro 4/3 and DSLR cameras are well suited to panoramas. Here's a simple work flow.
1. Set the camera to manual exposure at a point which best suits the panorama. Usually biasing the exposure a little darker if anything, as it's easier to pull out shadow detail than to recover highlights.
2. Manual focus, again to a point which best suits the whole image.
3. Shoot a 'set' of images overlapping each new image with 1/3rd of the previous - so the software has plenty of overlapped image area to work with - all with or without a tripod, though hand held you need to be shooting a distant scene and be very aware of the horizon, ensuring the images are all at the same 'elevation'. You can process the images to optimise them - as long as you do the same for each image.

Here, I have used a Nikkor 17-35mm lens set at 24mm, and shot about 6 frames in landscape mode - hand held. With the greater dynamic range of the DSLR (Nikon D2X), I was able to maintain highlight and shadow detail. This image looking back towards Cortina d'Ampezzo from above the Falzarego pass is 13300 pixels wide.

Cortina-06-0771 When the closest part of the image is less than 5-10 metres away, the proper method is to use a panoramic head on the tripod, which in its simplest form just ensures that the camera rotates around the 'nodal point' of the lens. This means that each part of the image - whether it's close to, or further from the camera - doesn't move in relation to each other regardless of the camera rotation - which would otherwise cause problems with registration between the images. Panoramic heads come in various guises and some allow stitching of more than one row of images (like the Agno one below. The various controls are there to get the lens's nodal point over the axis of the tripod head, about which the camera is rotated horizontally - for single row panoramas, and vertically as well, for multi row panoramas. As alluded to above, for views where parts of the image are close to the camera, it is essential that this is set up accurately. The nodal point can be found by swinging the camera on the head whist viewing a close (say 30 cm) and distant object through the viewfinder. The camera is moved back and forward along the rail (behind the camera base in the shot below) until the objects are stationary in relation to each other whilst the camera is rotated.


Here are two images; one shot with the panoramic head and a cylindrical projection, the other with an ultra wide lens (11mm on a Nikon D300) - 17mm equivalent full frame). 

Conquest Office

Conquest 1644-1

Image stitching
You need Software to stitch the images together. There are quite a number of options out there with varying amounts of control and complexity. I just use Adobe Photoshop, which is works remarkably well. Photoshop has a number of options to give various amounts of control over how it stitches images. I tend to use the cylindrical option which for me is the most useful for really wide panoramas. The weakness with the Photoshop software, is in its total automation of the stitching process with this profile, and if it can't 'match' enough visually identifiable points within the overlapping images it can't stitch. So images with only blue sky in the top half of the image. Some of the other software options give much more control over difficult to stitch images.

As I use it mainly for interiors, this isn't usually an issue. You can stitch either to create a virtual cylinder - which allows you to go right round to the full 360˚, or just to create a wider view with a perspective 'corrected' image. This last option will allow a much more limited angle of view, but keeps straight lines straight, which the cylindrical option doesn't.

This is a cylindrical projection covering about 170˚ with a panoramic head and a 21mm lens (full frame) in portrait mode, made with about 6 images. Underneath -  in B/W to illustrate the different views one can achieve from the same viewpoint, one shot at 14mm, and then with an 8mm fisheye (DX 1.5 crop which would be equivalent to a 12mm full frame fisheye. You can see the different fields of view.


bathroom 14-24


Going wider with a tilt shift lens - if you happen to have one!
This last option - for me - used to be about getting higher resolution images from my Nikon D2X and D300 12 megapixel cameras for shooting interiors. It is perhaps a bit of a niche but here it is:

I have the excellent 11-16mm Tokina wide zoom, and I can get very sharp images with it. For an A3 bleed image which is needed at times, this is really pushing things to the limit for sharpness on an interior image with just 12 meagapixels (4000 pixels across) for high quality print. One option would be to use a narrower angle lens in portrait format, and then shoot a panorama with the panorama head, and stitch the images using the 'perspective' mode in Photoshop to keep the straight lines - straight. The problem is that the edges of the image have to be stretched and enlarged so much that you get too much quality loss - and stitching such a radically wide image into a correct perspective is not easy.

My solution, as I have the Nikkor 24mm f3.5 PCE perspective control lens, is to use its shift capability, so I can use all of its extensive field of view in one image, by shooting three frames across its field of view (overall, it has about he same field of view as a 17mm lens on a full frame camera), one shot with the lens shifted to the left, one in the centre and then one to the right. As long as I also shift the camera in the opposite direction by the same amount for each shot, (so keeping the lens in exactly the same position each time - see comment about nodal point and Panorama heads above), I get perfectly registered images which stitch without even using the panorama function in Photoshop if necessary. This is far more practical and reliable than using the panorama head and 'correcting' for perspective with software, and is achieved with a slider between the camera and tripod.

shift lens pano

The images below show the 22mm (11mm each way) sliding scale for shifting the lens - in this case in portrait format, and below of the camera, the "really Right Stuff' camera mount which allows me to slide the camera, the opposite way to the lens shift.

Pano-head a

This image shot with Nikkor 24mm f3.5 PCE lens on Nikon D300. three stitched images.  6300 pixels wide

Conquest shoot. Pink Bedroom and Green-Orange Kitchen

I now have a Nikon D800 with 36 megapixels (7300 pixels wide). My 24mm PCE is effectively 50% wider than on my D2X and D300, and I now have the Nikkor 14-14mm ultra wide zoom, so is there any need to use this technique for added resolution? I'm doing tests with the 24mm PCE and the stitching technique described, which will make my images about 70 megapixels (9600 x 7300) pixels) and with about a 17mm wide angle view, though as it is a squarer image, this is only an approximate equivalent.

70 Megapixel test panorama with Nikkor 24mm PCE and Nikon D800, with 3 images shot in portrait format and stitched automatically with photoshop. With a slider under the camera, this takes about 30 seconds to shoot - with care! Below, a 100% screen shot from mid left in the image.

24pce 3images

24pce 3images close

Shooting with this technique for extra resolution, just adds an extra dimension to the possibilities. The 24mm PCE is a great lens, well corrected for perspective, and contrasty and very sharp in the centre. It is a bit softer towards the corners than the 14-24, but in real world shooting, this isn't an issue in my experience, the extra resolution for really large images may just make it worth the effort!


No comments posted.

January February March April May June July (1) August (1) September (1) October (1) November December
January February (2) March April (1) May June July (1) August September October November December
January February March April May June July August September October November (1) December
January February March April May June July August September October November December (1)
January February March April May June July August September October November December
January February March April May June July August September October November December
January February March April May June July August September October November December