Behind The Lens: Thoughts On Stitching

CP Freight Hwy. 1/1A Junction 2 blend, two stitch

There's already a how to article on my website on stitching, but I thought I'd discuss the whether to stitch issue. The image above illustrates some of the points. What you see above is a fairly wide shot - ideal for stitching - but it includes a moving object - no good - except that it's only in one half of the image - we're ok. Now, to complicate things, the frost was really great when I arrived but was disappearing while I waited. Fortunately I did a two image stitch after arriving, then took a third image with the train, blended the train with the right first image and blended the two with the left first image and voila. Of course you can't always get this lucky.

6 image stitch

Some things generally don't lend themselves to stitching - waves definitely comes to mind. Waterfalls aren't a problem so long as there is a decent overlap in images so you have some control over where the stitch occurs. The second image is a six image stitch - 2 columns, three rows with about a 50% overlap - from a Sony 707 and makes a quite nice 22X22 inch print. Borders between images had to follow lines of water, not across them.

I'm probably not the only person drooling over the incredible detail the new P45 and other similar medium format digital backs produces, yet at $40,000 cannot even imagine owning one unless I win the lottery (and I don't buy tickets). It's generally agreed that even good 35 mm. format lenses aren't up to any more pixels than the current 16 of the Canon 1Ds2. Think about it though. When stitching, you use a longer focal length lens (which is just as sharp) and by 'getting closer' this longer lens sees more detail than a similar resolution shorter lens (like using binoculars to look at a bird). So, when you multiply the focal length of the lens by 2 to crop to a partial image, which is then moved and repeated for other sections of the image, you get exactly twice the resolution of the ultimate image - cool huh? And remember, this is twice linearly, so that means the equivalent of 4 times as many pixels, each resolving as sharply as the original pixel.

Here's the math, just to confirm this:

16 MP camera, 5000X3300 pixels for a horizontal shot, now rotate the camera 90 degrees and zoom in or change lenses by a focal length increase of 5000/3300 or 1.5 to get the same field of view vertically as before. Now shoot 3 images overlapped to cover the old width of the original single image. This gives you 5000X7500 pixels =37.5 Megapixels which is almost the same as the P45 back. Remember though that you increased the image resolution by a factor of 1.5 by increasing focal length so you not only have the pixels of the P45, you have the resolution of the absolute best lenses around.

Now, depth of field changes too - since you are shooting longer focal lengths you have less depth of field so just like a regular medium format camera, you don't have the depth of field of 35 mm. That's the downside of stitching. Mind you, since Canon make shift and tilt lenses (24, 45, and 90 mm.) arguably you don't need to purchase both a Linhof 679 and a Hasselblad H2 as Michael Reichman did. Of course there are advantages to his system (apart from selling your house) for shooting people, waves, moving vehicles, moving machinery, and also for just plain convenience (you don't have to stitch).

I find shift stitching with my Canon TS-E lenses ok except that tiny changes in alignment as I move the camera over on my tripod head (Arca-Swiss) means that alignment isn't perfect and I still need to use a proper stitching programme. I can't simply bring the images into photoshop.

The two biggest problems for me with stitching are:

1) when you aim your camera up or down with your ball head, you have to either guess the nodal point, or you have to purchase a large, heavy and very expensive 3 dimensional stitching adapter for your tripod. Both Really Right Stuff and Bogen/Manfrotto make these devices.

2) and perhaps the biggest issue for me, in reviewing the images for consideration of printing, you can't pick out the winners without first stitching them - which means that it can be a year or more before I decide that a particular series of images will stitch together to make something worth while. I do sometimes let photoshop stitch the image but even that takes time and the temptation is to keep the Photoshop stitch but I absolutely cannot recommend that - the stitches are almost always flawed, don't align and often show seams. The simple solution is to shoot an initial single image, but this sometimes involves changing lenses, and I don't know who thought landscape photography was slow - for me things change way too fast - clouds move position, the wind picks up, the light changes. I don't want to blow the stitch by fumbling with a different lens first. I'm thinking that I might get a small light digital consumer camera with a big screen, put it on black and white and use it for viewing and for taking that initial image. Then of course I'd have to somehow relate the stitching images to this single image from a different camera - hmmn, not as handy as I thought. Back to the drawing board!

Now you know why most people aren't willing to stitch. Absolutely no question though - stitching with my previous 10D has allowed me to make some lovely large images. I recently restitched my columbia icefields image and made a 15X60 print that is very nice even up close. I confess that with the 1Ds2 I do get lazier and don't always use multiple images since it produces a very nice 16X20 but for bigger prints...