To make a complete 360 x 180° panorama with -only- eight photos from a 14 mm rectilinear lens?

Yes, we can!

Yes, we can and it's easy to do. Then should we?

Motivation

Of course you would guess that this shall be only possible with a full frame camera. Then, the diagonal field of view is about 114 degrees with such a wide-angle lens. This translates to about 104 and 82 degrees along the two axes (i.e. measured in landscape and portrait mode respectively).

There are many 14 mm lenses that are currently available from the major manufacturers. Much less are compatible with a full frame DSLR though. Some are zoom lenses: for example Nikon has discontinued the production of many older ultra-wide-angle lenses and replaced the 14 mm f2.8 by the 14-24 mm in 2007 (to fit with the FX new sensor).

Some more lenses with this short focal length from major independent lens producers are of an excellent quality and compatible with most of the different FF camera mounts. In 2010, Samyang introduced the 14mm f/2.8 IF ED AS that I own and have issued a review about it. When concluding this review, I wrote: This lens is excellent for stitching panorama photographs though. But the panorama output resolution advantage is rather meager when the number of images to make a complete 360 x 180 panorama is taken into account. A 15 -16 -17 mm fisheye lens is IMO better suited for "exclusive" panorama photographing. No less than 12 images on a FF camera are needed to close a spherical panorama. Adding one Nadir and one Zenith to make 14 images in total is probably better for outputting a 13600 x 6800 pix.

Well I was wrong: it is perfectly feasible to stitch only 8 (or better 9) images from this excellent lens. Six horizontal shots, one zenithal image and one (or better two) nadir image can be stitched to complete the whole object space. That's now a different new game all together!

This article is then to correct my peremptory pre-conceived point of view.

Foreword (important)

All the examples (photos, VR panoramas and drawings that are shown on this page were obtained by using souce images shot with a Canon EOS5D2 camerathat was mounted as usual on the upper part of the Nodal Ninja 5. However a home-made "un-precise" Slim Rotator and plate were installed to replaced the horizontal rail and the RD-16 rotator of the NN5.

This Slim Rotator is very precise as alignment is concerned, but the angular spacing is not: there is up to 4 degrees random variation in the angle between successive shots, therefore the test that were performed are fully representative of a low-end rotator. I have designed and build "Precise" version of the Slim Rotator but I deliberatly did not use those. Bad results could also certainly have been avoided by using more accurate commercial gears such as the NN5 itself, but this change was done on purpose to test the (6 + 1+ 1) in a worst type of conditions.

The technique 6+1+1 is applicable in any occasion where a 15-16-17 mm fisheye could be appropriatly used i.e. for outdoors, indors, in tight space and close-up panoramas. The images from the Samyang lens are almost totally free from lateral color (aka chromatic aberration) and consequently the following panoramas were stitched from JPEG images downloaded straight from the camera (without any post-processing).

Examples of panoramas. (All done from the JPEG files straight saved in the Camera)

Inside a large building room: in a rural French church during Christmas season.

 

Landscape photography: during winter time, no barge needs to navigate on the Canal du Midi, the Locks of the wild boar have been emptied.

 

Landscape photography: early morning winter haze over Lauragais.

Panorama photography in a tight volume: interior of an automobile (LDR fusion from multiple exposures).

Shooting the source images

Compared with the usual situation when using a fisheye lens with same limited number of images, there shall be a rather smaller overlap between some of the six "horizontal" images and the zenith an with the nadir image(s) as well. In fact if not enough care is taken, some tiny holes might appear in the panorama.

With a minimum of precaution, there should not be any useful pixel missing:) . Here follows a list comparing the new to the usual way (i.e. with a fisheye F.E.):

  1. Use a good 2-axis panohead and rotator. The more precise is best but ultra-precision is not absolutely required. Naturally the camera shall be mounted in order to get photos in "portrait" mode, as usual. Not too much difference with a F.E.
  2. The "horizontal" plane is allowed to be "not exactly horizontal" but the axis of the lens should be perpendicular to the pivot axis when shooting six photos while aiming "horizontally". This latter requirement is more stringent than for F.E. Good panoramic head have a physical stop to ensure adequate and accurate perpendicularity.
  3. Shoot these six photos at 60 degrees interval. You should then better use the relevant specific setting on the rotator. Same as for 15 or 16 mm F.E.
  4. Zenith and nadir shootings require that the lens is aiming straight up or down with respect to the "horizontal plane". Good panoramic heads have also physical stops to ensure enough accuracy in that matter. The F.E. can be willingly let aiming away from true Nadir or Zenith. No such option exists with a 14 mm rectilinear. That's quite a difference here.
  5. Two nadir shots spread 180 degrees apart allow to get rid of most of the panorama head obsruction at the final blending stage. This is a well-known trick and not specific to 14 mm shooting. Same as for F.E.

While not being an absolute requirement with the Samyang, the best is to shoot the Zenith and Nadir photos with the rotator set for 30 degrees steps and subsequently to put the rotator at an alternate position (that is to say between two click-stops that were previously used at step 3 above). Under this condition the overlap shall be nicely and evenly distributed. The stitching program (Ptgui, AutoPanoPro, etc.) shall then generally and easily find control points and shall auto-align the whole pano. On the contrary, if the same rotator position (as the one that was previously used at step 3 above) was selected, then in some place the overlap is "nearly" nil at very specific location of the equirectangular panorama as it is detailed in the next chapter.

 

Why is that possible?

The angle of view of this lens on a full-frame camera is sufficient to cover the whole space with eight images 6+1Z+1N; With a second nadir shot at 180° yaw from the first one, most of the panorama headobstruction is wiped off.

Caveat lector:

Lets suppose that the Zenith of the panorama is as follows:

If the orientation of the Zenith (yaw setting on the vertical rotator of the panohead) was shot is the same click stop position as one of the shots in the horizontal plane, the result is illustrated below:

 

or in a simplified illustration:

One can see that the overlap is scrce and twoplaces a high risk of a lack of image exist ( Red warning signals). A precise panorama head and rotator with accurate and equal angle spacing are required to make sure. In consequence this process cannot be recommended.

or on a simplified illustration:

When the zenith shot is oriented halfway in order to bisect anywhere in between two shots of the horizontal row of 6 images then the situation is totally different: There is ample margin and no risk to have a gap with only useless black image pixels. Low-end (or home made) panorama head can be safely with this scheme.

Overlap

As the zenith image is concerned the overlap area with the images of the horizontal plane is about 64% total. Four images (1, 3, 4and 6) have about 13% common with the zenith image.

What would it be like if it was a 15 mm lens instead of the 14 mm lens?

Whis a 15 mm lens, it is impossible to make sure that no gap would appear on the Zenth or on the Nadir. There follows an exemple of a Zenith of a panorama that was shot with the same camera than above (EOS 5D Mark II) and a Nikkor 15 mm f3.5 (circa 1972):

The margin of the Zenth image and two lateral images is now nil!

It's obvious that doesn't work.

 

What would it be like if it was a 17mm (Full-Frame) fisheye lens instead of the 14 mm rectilinear lens?

There is then plenty of overlap and the stitchers are at ease with this configuration. Of course. The Tokina 10-17 mm @ 17mm that was used here for testing gives notably softer output panorama image than the Samyang.

Plenty of overlap in this case!

Side note: Improvement of panorama VR technologies in about six years (2004 - 2011)

It is impossible to objectively compare technologies from different eras because too many factors should be accounted for, but anyhow I shall put the same panorama that I shot in July 2004, in September 2005 and in February 2011 side by side. Beware: the latter is quite big and multi-resolution.

While the first was based on four images shot with my first DSLR i.e. the EOS 300D a.k.a. Elan (APS-C; 6 MPix), the second was done with a Canon EOS 20D (APS-C; 8 MPix) and the same Sigma 8 mm f4.0 (Fisheye vFOV= 183°) to stitch four single images and finally the third was done recently as was described in length above in this page with a Canon EOS 5DMark2 (FF; 21MPix) and a Samyang prime lens 14 mm f2.8. (vFOV= 105°) to stitch 9 x 5 images "enfused" from 5 +/- 1.7 EV bracketed photos;

The native résolution of the equirectangular output is more than doubled i.e. 4X more pixels @ (13500 x 6750) Vs (6750 x 3375) Vs (5650 x 2825) and the noise level has been dramatically decreased. Following are a screen capture of the Zenith of the three panoramas at approximately the same viewing zoom level:

EOS 300D + Sigma 8 mm f4.0 + PTMac 2.0
EOS20D + Sigma 8 mm f 4.0 + PTGui 4.1
EOS5DMk2 + Samyang 14 mm f 2.8 + EnfuseGui + PTGui 9.0.2
July 2004
September 2005
February 2011

What shall the same panoramas look like in 2016?

The more recent panorama endeavor above on the right was the subject of a test of several panorama software: I wanted to know their respective ability in handling this challenging method that pretends to make a very high resolution panorama with the minimum number of shots from a rectilinear lens (not a fisheye, mind you).

Conclusion:

A 15 mm or a 16 mm fisheye lens and a full frame camera was known for long to be the best combo for shooting high quality and higher resolution 360 x 180° panorama while having a minimum of images to shoot, ( 6+ 1Z+ 2N) being the de-facto standard shooting scheme. A similar but not resolution equivalent solution is to use an APS-C camera with a 10 mm to ~12 mm fisheye or even possibly 8 mm (depending on th F.E projection type).

I have recently discovered that in fact this is also possible with a 14 mm rectilinear lens on a FF camera. Taking into account the outstanding IQ of some of these lenses (i.e. the Nikkor AF-S 14-24 mm f/2.8G ED FX, the Canon EF 14 mm f/2.8 USM L II and the Samyang 14mm f/2.8 IF ED UMC AS for example), they are more able to produce superior panorama quality because of the intrinsic lesser need for interpolation during all the final step of panorama stitching.

Further more, these lenses have been almost perfectly corrected from chromatic aberration. BTW this is particularily true for the Samyang. As a result of trade-off to get this outstanding features the radial distortion is let not very well corrected (to say the least when Nikkor and Samyang are concerned), but that is of no consequence for panorama stitching ability. When the exposure is good, there is almost no compulsory need remaining to convert the raw images for stitching panoramas. Using the JPG straigh out from the camera is a fast and -at last- acceptable way since vignetting can be corrected nicely by the stitching SW...

The Samyang 14 mm costs about 6 times less that the others but it gets full manual control only (on Canon). IMO It's a real bargain.

Therefore I shall change my mind: that focal length shall be my favorite panorama stitching photographic tool on the Canon EOS 5DII when high Image Quality is required. The Tokina 10-17 mm @ 12 mm shall complement it when singl- row limited quality is sufficient (most of the case when viewed on-line).

 

Michel Thoby

23 January 2011

rev 26 Feb 2011 (Comparison 2011 Vs 2005)