By: Bernard Languillier
Image quality... Although I am convinced that the essence of top level landscape photography lies elsewhere, there is no questioning the impact of a very large sharp print.
A fascinating comparison of high-end digital backs was recently published at The Luminous Landscape. The conclusion that struck me most was that a casually processed 39MP file is real close to a very well scanned 4X5 transparency.
Is there a way for rest of us to get close to these wonders without selling our sailing boat?
There is. It is called stitching. Quoting somebody whose name I forgot, – "Stitching is the Linux of image quality," – a mostly dirt cheap method to get sky wide prints with huge potential but some limitations, just like Linux in a way.
Stitching has been around for some time. The most commonly practiced method is spherical stitching. The camera is rotated around the entrance pupil of the lens so as to record a facetted cylindrical or spherical image. The individual images are then blended and projected to a plane using a specialized piece of software. Recent advances have made spherical stitching a breeze. High quality spherical pano heads from vendors like RRS are now readily available. Very cheap software packages like PTgui 5.8 are now able to automatically stitch most images and have reached a very high level of quality. I am an addicted spherical stitcher.
There is more to stitching than spherical stitching though. Keeping the lens in a fixed location while moving the sensor makes it possible to do flat stitching without the need to do any cylindrical projection. Since the images all belong to the image plane, opening them in PS is often enough to get a perfect stitch by doing a straightforward overlay. Software like Realviz Stitcher 5.1 can further automate this process.
Some people have been using shift lenses in combination with L brackets in order to do this. Overlaying 2-3 images with this method basically doubles the resolution of a DSLR at low cost, at least if you are lucky enough to be a Canon or Nikon owner. Light fall-off can be pretty bad though.
Building further on this basic idea, the large format camera manufacturer Horseman released early 2006 a model called Horseman LD that was designed specifically to bring the flexibility of LF shooting to DSLR owners, including back movements like shift.
1. It comes with a Canon EOS or Nikon F lens mount so that mounting a DSLR is straightforward
2. All the movements of a LF view camera are theoretically available (more on this below), including front and back movements,
3. Flat stitching becomes possible on a much larger scale than when using a normal DSLR T/S lens.
After having spent more time recently playing with the Horseman LD with my D2x, I have compiled some first feedbacks in the following chapters. I have also tried to do a very casual comparison with spherical stitching and 6X17 scanned film on a given scene.
As a disclaimer, I am still very new to the LD, and the present article should not be seen as a full in depth review. I urge people interested in purchasing one to test it first to confirm whether it meets their needs and expectations.
The unit might not be up to Arca Swiss or Linhof standards, but is well built and as a consequence pretty heavy. According to Horseman, it weights about 4 kg.
Fig. 1 : Basic LD set up
Fig. 2: Basic LD set up
The DSLR is attached to a lens mount located at the end of a special bracket sticking out from the rear standard. The mount can rotate about 100 degrees so as to enable the rotation of the DSLR from horizontal to vertical mode.
Fig. 3: F mount on rear standard
The LD is rather bulky, and I would find it pretty challenging to pack the LD in my usual backpack for trekking without un-mounting the standards. Need to think more on this.
Both MF (Mamiya or Hasselblad) and LF lenses can be mounted on the front standard thanks to Horseman adapters.
MF lenses do mount on a well designed special recessed board.
Fig. 4: Mamiya 45 f2.8 on LD
LF lenses mounted on a Linhof board can be used as such thanks to a reduction Sinar -> Linhof adapter.
Fig. 5: Schneider 110 f5.6 on LD
All movements are available on both rear and front standards. They are all geared for accurate and repeatable positioning. The neutral positions are marked by a stop that makes it rather easy to calibrate the LD back to its neutral position. The design appears to be very similar to the classical Horseman rail cameras, except for the rear standard that was modified to enable the mounting of a DSLR.
It is possible to fine tune the vertical mounting location of the DSLR relative to the rotation axis of the rear standard. This will come handy for those willing to use back tilt, since it should be possible to simulate Sinar/Ebony like asymmetrical back tilt. I havn't reached that level of sophistication yet though. One should not forget to adjust the front standard neutral position accordingly, in order to preserve the alignment of the lens and sensor centers.
Both front and rear standards come equipped with a bubble level that help adjust the front-rear horizontality of the camera. There are no left-right levels though but I have found using a DSLR flash shoe mounted 2 axis bubble level to be a suitable by-pass.
The Horseman LD appears to cost about $2500 USD in North America. It is significantly cheaper in Japan since I paid only about 200.000 Yen for mine a few months back, which now corresponds to $1,600.
My first hours of shooting with the LD left an overall positive impression. Things work per the plan but there are a few limitations and issues to be aware of.
Although the front and read standards are correctly calibrated vertically, it appears that the sensor of the D2x is not centered with a lens for which the front standard would have been set a zero (no lateral shift). There is a need to apply a 8-10 mm side shift to bring the lens center in alignment with the sensor center. Without doing this, vignetting appears faster on one side than on the other when doing lateral back shift.
Perfectly aligning the DSLR with the standard isn't ft as easy as it seems, but this is key for easy stitching.
Fig. 6: Horizontality check
When working with a lens like the Mamiya 45 f2.8, the applicable amount of tilt is very limited because of how close the DSLR is from the front standard. Typically only +/- 5 degrees in tilt is possible. Down looking tilt of the rear standard (very rarely used for landscape) is almost impossible (2.5 degrees only).
Since there is of course no coupling between the lens and the body, you'll need to stop down the lens manually before shooting. Pretty usual stuff for LF shooters, though easy to forget when you shoot with a DSLR.
There is a need to focus MF lenses near infinity to be able to focus the whole thing at all, at least for wide lenses like the Mamiya 45 f2.8. Since the distance between the prism of the D2x and the front standard is only 2-3 mm when focused at infinity, it is better to adjust focus using the lens focusing ring instead of using standards movement. Indeed, it is very easy to move by mistake the rear standard too close to the front one, therefore causing the prism of the DSLR to touch the front standard. This would induce an unwanted slight tilt... This might in fact only happen when applying an upward shift to the rear standard... go figure why some of your flat stitched images are suddenly blurred... It took me a while to figure out myself.
It is difficult to use the D2x in portrait mode because of interference between the support bracket of the LD, and the cable release connector. Not a problem if you don't use a cable release (but that is kind of mandatory I feel).
Test Set Up
LF camera: Horseman LD with F mount adapter.
Lenses used: Mamiya 45 f2.8 for 645, and Schneider 110 mm XL 5.6 for LF.
Body used: Nikon D2x.
Software used: RSP Premium 1.02 for RAW conversions, PTgui 5.8 for spherical stitching and Realviz Stitcher 5.1 for flat stitching.
Achieving critical focus proved to be easier than I had anticipated. At least, the results appear sharp enough when the lens is stopped down to f11-f16.
I got interesting results with lateral shift. The subject isn't too exciting,
but this little river located next to my home was
convenient to take 15 kgs of gear to. With the Mamiya 45 f/2.8: up to
+/- 20 mm and vertical shift up to +/- 10 mm (perhaps even a bit more) could be achieved.SPACEThis results in a perfect stitch that is about 10.000 pixels wide by 4500 high from 10 landscape images. These 10 images can be taken very quickly, I would say in 10 * (exposure time + 2 sec). There must be some light fall off towards the corners of the scene, but I didn't notice any with my test subject.
The image shown in Fig. 8 was shot from RAW frames that were converted with RSP Premium 1.02 with no sharpening or detail extraction, but application of a slight curve and color vibrance. The stitch was made with Stitcher 5.1 (normal sharpen). It should be noted that a .5 degrees lateral back shift was applied to improve DoF.
Fig. 7: Stitched image
Fig. 8A: 100% crops image left part
Fig. 8B: 100% crops image central part
Fig. 8C: 100% crops image right part
With a Schneider 110 mm: there is even a bit more room for shift, but I don't have the exact figures. The difference is not so big, and mostly results from the difference in size and location of the lens rear elements, not its image circle from what I could see. The sample below was shot when 20 mm side shift was applied. Fig. 9B is a 100% crop. A bit soft, but still pretty acceptable I feel.
Fig. 9A: Schneider 110 f5.6 image quality
Fig. 9B: Schneider 110 f5.6 image quality 100% crop central part
With both of these lenses, the resulting images are sharp, probably not 100% as sharp as those I get from my D2x with good old Auto Focused Nikkors at f11, but close enough. I'll leave it to more knowledgeable readers to check this impression by themselves.
What happens if you over shift? Well, you get nasty vignetting. Fig. 10 below was taken with a 30 mm left shift.
Fig. 10: Vignetting sample
Interestingly, this vignetting isn't related to the image circle of the lens. It is induced by the lens mount and/or bellow. Fig. 11 shows what the sensor saw as I progressively shifted the D2x to the left:
Fig. 11: Initial capture
Image circle related light fall off would creep in towards the left end of the image and end up with a sharp cut off. It is clearly not what we have here. The problem with the behavior at hand, as shown in Fig. 12, is that the last frame ends up being mostly un-stitchable if there isn't enough overlap with its neighbor.
Fig. 12: Hard to stitch...
Using the movements of the LD to increase DoF by using the Scheimpflug rule is a lot more difficult than when working with 4*5 because of how small the DSLR viewfinder is. The impact of subtle movements is very difficult to assess without taking a picture and checking the sharpness at 100% on screen, which makes the process slow and cumbersome.
Combining flat stitching and Scheimpflug is possible, but even trickier to do. Since there are always elements in a scene that fall outside the plane of sharpness, most people proceed by iterating so as to find the best compromise between the level of blur of these elements. When working with 4X5, the whole scene can be checked readily on the ground glass, but it is obviously not the case when doing flat stitching with the LD. Shift will have to be applied first to check the result on the DSLR screen in other portions of the global stitched image. This makes the definition of the plane of sharpness extremely cumbersome at best.
A simpler scene, like a beach at sunset, would be easier to handle since each image making up the mosaic is representative of the requirements in terms of plane of sharpness. There might be smarter ways to proceed though and I still reserve my judgment on this topic.
Flat stiching using Stitcher 5.1 was OK, but a bug in the software prevented fully automated stitching. I had to use the semi automatic stitch mode, still fast, but not as fast as I would like.
Comparison with Spherical stitching and 6x17 scan
This is not a rigorous comparison, even if the point of view is approximately the same, the crops are rather different and the shooting parameters not exactly the same either.
For what it is worth, here are the results obtained with 2 additional shooting methods for the same kind of scenes.
Fig. 14: Really Right Stuff Spherical head
Lens used: Nikkor AF 35 mm f2.0 at f13.
Raw converter: RSP Premium 1.02 with no sharpening or detail extraction, but application of a slight curve and color vibrance (settings very similar to those used for the flat stitch images).
Pano software: PTgui 5.8 using Lancosz interpolation. PTgui doesn't provide detailed information about the sharpening that is applied.
Resulting image size: 2 rows of 5 images 8600 x 3200 pixels.
Fig. 15: Stitched image
Fig. 16A: 100% crops image left part
Fig. 16B: 100% crops image central part
Fig. 16C: 100% crops image right part
The spherical stitch images are very close to what we obtained flat stitching with the LD. Slight differences in framing and resulting image size make it hard to compare the 2 rigorously, but the results are definitely in the same ball park.
Fig. 17: Da Yi 6x17 120 film back on Ebony 45SU
Camera used: Ebony 45SU with Da Yi 6x17 film holder. A slight back lateral tilt was applied.
Lens used: Schneider 110 mm XL Super Symmar f5.6 at f22.
Film used: Provia 100F. Scanner used: Imacon Precision III with 6x17 holder at 3200 DPI.
Resulting image size: 20.900 * 6985 pixels (about 840 MB).
Fig. 18: Panorama image
Fig. 19A: 100% crops image left part
Fig. 19B: 100% crops image middle part
It should be noted that the crop of the 6x17 image is significantly wider than that of the stitched images above, which penalizes it to some extend. I decided to shoot from the same tripod location and the Schneider 110 mm was the closest lens in my arsenal.
The lack of contrast and saturation of the Provia 100F results mostly from a small internal reflection problem with the 6x17 back in the absence of a lens shade. Besides, the scan did not benefit from the enhancement of colors that RSP did on D2x images. This comparison is not meant to be a scientific test, and color accuracy is certainly not one of the aspects that should be compared here. The difference of contrast resulting from the internal reflexions will probably reduce the impression of sharpness of the scanned Provia 100F.
Put briefly, the amount of time and thinking spent on each of these shots was about the same, but the shear simplicity of the digital workflow made it a lot easier to achieve reasonably pleasing results, even if film is still a good contender in terms of details.
The Horseman LD has proven to be a credible solution for flat stitching. Besides, it should provide good value for architecture work with a DSLR, where rise capability is essential.
For landscape applications, I see some issues. First, I personally find it difficult to tap into the LD's Scheimpflug potential because of the difficulty of accessing the impact of a tilt on the sharpness of the image in a DSLR viewfinder. Besides, its sheer bulk and weight make it a bit difficult to take along on long treks. It should not be a problem for those who typically only walk short distances from their vehicle.
Overall, I still believe that cylindrical stitching with a good spherical head and PTgui is the best solution for landscape work today, although the cheap Da Yi 6x17 back has proven once more that film isn't dead yet.
Bernard Languillier is an amateur photographer based in Tokyo.
He focusses on landscape and cityscape work.
Some of his recent images can be seen at http://www.flickr.com/photos/bernardlanguillier/sets/
© 2006 Bernard Languiller