Story of a panoramic photograph
Tuesday, 18 November 2008
|
Written by Riccardo
This article in Italian
A bit of theory about multiple shooting panoramic photography.
Making a panoramic photograph is made possible today by latest technologies, with a minimal technical effort, time and cost. Only a few years ago this was a much more laborious task and required a costly specialized equipment such as camera with rotating lenses, affordable only to those who had the real need.
The typologies of wide-angle panoramic pictures are several, as are various the techniques to achieve them.
For the most part, they have in common the need to merge several shots to create a single image.
Although there are some exceptions. A couple of examples are represented by those photographs taken with a 180° FOV fisheye lens and by the already mentioned cameras equipped with special rotating lenses.
For simplicity sake, we can distinguish three broad families of panoramic photographs:
- Planar projection
- Cilindrical projection
- Spherical projection
Planar Projection Panoramic
A Panoramic planar projection is the most simple in its concept: it consists in capturing a subject with several shots, having care to keep film plans coplanar and parallel to the main plane of the subject. The focal length is the same for all shootings. The higher the focal the higher the accuracy you'll reach, decreasing sensibly the perception of parallax offset at the edges of each frame. On the other hand, a higher focal length will also require a greater number of shots.
Once you're in your studio you'll realize how easy is to merge the various images taken, even without the aid of a computer.
We can observe here below a simple scheme representing a series of captures, in a single horizontal row:


A similar technique is used to represent accurate aerofotogrammetric maps, from top to bottom using an aircraft.
Although the concept is indeed simple, the execution is not: one must prepare a detailed shooting plan in advance, then perform field surveys to properly position the camera to ensure a correct framing. The operations are further complicated in case you'll need additional horizontal series of captures: to position the camera mid-air is not an ordinary task.
Cylindrical Projection Panoramic
Albeit a Cylindrical Projection Panoramic could be conceptually a bit more challenging, it can be executed with relative ease: a series of shots in a circle - approximately centered in the camera lens - capture the scene around over a cylinder (real if using lens rotating cameras, or virtual and discretized if using conventional cameras to take a continuous series of shots). This cylinder will be then developed (unrolled) on a plane to obtain a flat print. For this type of photographs there's no claim for a photogrammetric accuracy.
The maximum angle of field we can cover is obviously 360°, although we can just limit the entire set of capture to a lower range of degrees values.
Here below we can observe a schematic view of a set of captures, one single circular row:

Theoretically, the center of the rotation should coincide with the optic center of the used lens system. To maintain the explanation the clearer the possible, right now we don't overly concern about the exact center until we don't plan to take shots including objects positioned very close to the camera: that's the case where an excessive parallax offset would emerge. To successfully manage very close subjects (typically in interior spaces) or entire foreground layers eventually framed would be better to use a specific decentrable panoramic head mounted on tripod.
It is now possible to extend the capture towards the top and bottom by adding other circular strips, tilting the camera some degrees:

By tilting the camera the contents of the frame will be projected on the cylinder and two different side effects will be introduced: a gradual upward stretching of the image and a blending of formerly straight lines: this is clearly visible in the diagram below - green contour: the frame is not distorted (or a just minimum), red contour: frame distorted -
Note: For a better understanding and explanation of the graph, the portion of the cylinder is represented before the optical center (where the projection rays converge). Ideally, the portion that matches, or more precisely, that finds off on tangent to the focal planes of the individual shots, is exactly where the image will be projected. Having care to maintain the axis of revolution of the cylinder coincident with the axis of our rotation - the optical center - the represented geometric proportions do not change, except for the fact that the image will be projected upside down.

The more the camera will be tilted the more the distortion will be evident. A zenithal shot projects the image to the infinite. It is necessary then limit the tilt angle of the camera to low values.
Spherical Projection Panoramic
A workaround to this problem could be a projection of the frame on a spherical surface to keep unaltered the proportions of the objects represented. Then, the use of a software to develop the spherical projection on a plane. It would be not an easy task, especially on the field, and, all in all it would be not very interesting if your purpose is a traditional planar print: You would see the ground at your feet and the feet themselves! at the same time you would frame an excessive portion of the sky. This Technique is instead suitable to reproduce a navigable virtual spherical environment on screen.
A Case Study: The Panorama of New York City
The goal of this article is to obtain a landscape photograph with a cylindrical projection with an angle of field equal to 360° over the horizontal plane, in the most simple and effective way. Making the task within every one's reach. And in a reasonable short time.
The projection, however, will have a field angle on the vertical plane equal or less than 100° or, as already mentioned above, the stretching of the upper areas would be excessive.
The equipment we do need:
- A camera with a lens with a focal range between 20mm and 35mm . For best results use a reflex.
- A tripod (optional but recommended, in the example of this article no tripod was used).
- A computer and a software (photoshop or photoshop elements in this case). Memory amount really depends on the resolution of the final image.
- A scanner, just if use a film camera.
On the field:
Let's choose our subject to capture. I was positioned in the far southern tip of Roosevelt Island, in East Hudson River: from that point there's a good view of the city of New York: East Side, Midtown, Turtle Bay, East Harlem, Queens and Long Island City, Williamsburg Bridge, Queensboro Bridge, Downtown, many significant buildings and of course the island itself.
The skyline of New York is a subject photographed billion times. The reason resides in its undoubtedly spectacularity and magnificence. That's the typical situation when it becomes really difficult to photograph something that was not already seen or something that doesn't bore.
I was looking for a spot from where I could capture a view of the city. A view that I had never observed before in other works and that were, at the same time, effective. The idea of making a panoramic view raised in my mind after this process, as a complement of the inspiration to photograph the city from that point.
I waited for the right weather situation: the sky had to contribute significantly to the composition.
Look for a subject that looks interesting to you and suitable for a panoramic, avoid to reverse those actions. Circumscribe your technical skills at the service of your ideas. Doing the opposite often leads to dull results.
Acting the right way on the field is essential for a successful result. This rule always applies and it does a lot easier the next task, at home.
In this case, once you have chosen the right spot, start taking a series of photographs in a circle, extending the rotation enough to capture everything you want to include in the final photograph. If you like, you can cover the whole horizon circle.
As already mentioned, if you plan to frame objects positioned very close to the camera, pivoting approximately on the mounting screw of the camera (on a tripod) will be not enough to achieve a precise capture, without an excessive parallax offset error. Then, in case in your frame you have e.g. a close tree, a panoramic tripod with a decentrable head would be needed: make the rotation around the optic center of the lens system (the first nodal point, which indeed is not easy to geometrically identify , as a generic reference it's positioned about at half objective length).
-
Follow these easy recommendations:
- keep the camera horizontal.
may be useful a viewfinder equipped with a grid. If not available with your camera use the focus points as a reference. The use of a built-in bubble is essential to level-up your shots in those cases when the horizon is not clearly visible or totally hidden. - Use a focal length moderately wide (between 28mm and 50mm) and keep it strictly fixed along all series of shots. The more narrow the focal will be, the higher number of shot will needed to complete the whole rotation.
- Always keep a margin of overlap between one frame and the other 20% is the minimum. If you don't do it, all shots will be unusable. At the same time avoid to frame the same portion of scene more than two times, with excessive overlapping.
- Adjust the exposure at an average value that works out for the entire set of shots. Once you find the right exposure (the use of built-in camera's exposimeter should be fine), keep that value fixed. In case of excessive light contrasts, you can try a moderate variation of the exposure, keeping in mind the software could balance between a shot and the other. This applies only if the difference in exposure is really slight. Of course if you have a particular or non canonical idea about how to expose the scene, just do it. In my example, for instance, I kept in mind that the final photo had to be monochromatic.
During the captures it's probable you'll have to point at the sun: NEVER stare directly at it with your eye fitted in the viewfinder. It's absolutely not recommendable even naked eyes, now, imagine doing so through a lens! Just think about that's not impossible, under these conditions, to burn or melt parts of the mirror box...
We can now choose whether to perform two additional stripes towards the top and bottom to extend the field vertically, or just keep one narrow single strip. The tripod well positioned and the head mount leveled up (use the built-in bubble) are the key points to maintain the tilting angle constant. Keep an overlapping margin both horizontally and vertically between the shootings.
To cover 360 degrees using a 28mm will need about 8-10 shot per strip . So, for three horizontal stripes, we'll produce around thirty shots.
If you have no other chances to get back to the shooting spot, perform at least 3 series of captures for a total of 90 shots. Between a series and another change some parameters such as exposure, percentage of overlapping, etc..
It's very important to take the shots with a legible and logical order. Proceed and choose a rigorous method. If you use a digital camera name each series of shots. Otherwise using film, proceed with paper and pen. There's nothing worse than having many messed up good shots.
Once we finished, we can leave the field and go back home to elaborate the shootings.
We have to choose a software with advanced merging features. Thanks to its ease of use, I chose Photoshop or the cheaper Photoshop Elements. Both are not properly specialized stitching softwares. On the other hand, the objective of this guide is to achieve good results in a few moves. Working with more care and precision in the field and thereby saving time sitting in front of a computer. I also report a free and open solution that offers a comprehensive set of tools: Panorama Tools. However, a copy of Photoshop is available for one month as a free trial directly from the Adobe website.
Let's see what I produced for the guide example: These are the shots of three complete stripes as I loaded them on the disk:

I shooted handheld (easily comprehensible directly from my own shadow!) taking care of the aforementioned directions. I used a Nikon D200 (any other reflex is fine) and a 18mm (equivalent to a 28mm on 35mm film). Fixed exposure measured between the sky and ground with a slight prevalence on the sky in order to avoid washed-out clouds. Each individual image measures approximately 10Mpx.
It is therefore necessary first reduce and compact the images: 1600x1066 pixels is a reasonable value and needs no more than 2 GB of memory for a smooth elaboration. Once the assembling phase is terminated we will have produced a photo of around 20Mpx (including the final trimmings): a size useful for prints larger than 80 cm / 32" of width with no loss of detail.
800x533px tiles will serve a final image of about 5Mpx. Do some preliminary tests at 400x268px to make your work flow smoother, then, once you could see everything goes in the right direction, rebuild the image with a higher resolution.
To change the size of entire groups of pictures comes in handy a Photoshop feature: batch, under the file menu-> automate.
Put resized images in a folder, sorted by shooting order.
Launch Photoshop and in top menu, choose File-> automate-> photomerge
In photomerge dialog box check "auto" in layout options.
This option is try to produce a panoramic cylindrical projection without any further inputs, except for the image tiles:
explore the folder previously prepared and select the files from the first shots row, the horizontal one taken with the camera angle leveled up to the horizon, press OK and let the computer make its calculations.
We are doing some simple preliminary tests in order to achieve a better understanding about how the software processes and merges your images:
here is the result for the first row:

This is already a 360° complete panoramic view. However, we may want to extend the field of view vertically, to include parts of the scene as of the sky. We also might a more balanced format, decreasing width/height ratio.
In the same way as above then, we build the second row, the one taken tilting the camera upwards:

Here it comes the sky, a crucial component of this composition.
We observe while the first image is uniformly stictched along its entire width, in the second image we see some tiled images are higher or lower than the others: This is the result for using the camera handheld and inadvertitely slightly change tilting between one shot and the other. As already described above, use a tripod for a uniform result.
Further we see pronounced curved top and bottom boundaries and trapezoid shapes of the singles tiles images.
Last, I shifted the content of assembled image to the left, in fact the highest tower (Trump Tower) is no longer in the center of the frame. This can be easily done as, in a 360° panoramic image, far left and right edges perfectly matches. The shift helps us to better understand the progression of the lines that originally were straight (in-line clouds and airplanes lines),
Once observed this series of characteristics we can inspect the causes in the box below:
Why do straight lines are represented as curved?
Thanks to the type of projection (cylindrical) which has been used, the horizon is a straight line across the length of the frames. That's because in the reality it is a circle that we followed with our series of shots and it matches the horizontal axis of the first series of leveled shots.
All the other lines that are straight in the reality appear more or less curved, depending on their extention, distance from the camera and from the horizon. This happens thanks to two concurrent factors:
Thanks to the type of projection (cylindrical) which has been used, the horizon is a straight line across the length of the frames. That's because in the reality it is a circle that we followed with our series of shots and it matches the horizontal axis of the first series of leveled shots.
All the other lines that are straight in the reality appear more or less curved, depending on their extention, distance from the camera and from the horizon. This happens thanks to two concurrent factors:
Notice how the horizon behind the city skyline is properly represented as a straight and horizontal line, since it has been captured inline with the horizontal axis of the first shooting set. Once verified the goodness of the entire merging process we can close the image without saving it and repeat the operations, including now all the images from the three stripes set. Here's the result:

Corrections on Photoshop automerge function are not an easy task, altough if we previously made a good job on the field, we won't get in troubles. However, Photoshop puts on separate layers the single landscape portions with their cropping masks. For those who have some pratice in handling masks, it will be easy to recover any minor glitch. Once flattened all the layers you can always make some adjustments with the clone stamp or free transformation on small areas of the image. In the example, post processing and retouching were minimal, only the railing in the foreground requested additional work, since it was "not digested" along the main calculation. (See next image). A parallax offset error arose since the object lays in the close foreground. As already explained, the use of a decentrable pano head mounted on tripod most likely would have prevented the issue.
Be careful at both right and left sides when you're composing a 360° capture: if the image lacks of some content close to one border it is likely displayed at the other border. In this case, make a shaped cut and paste it from one side to the other, taking care to put the cut content on an independent layer during the repositioning. However, if different exposure at borders is excessive you can run a second photomerge pass between the main image and the side cropping. This time use the manual repositioning option (reposition).
You can now proceed with the cropping of the image to eliminate the black areas at borders. Inevitably, some content also will be cut off:

On the image above I made a slight color correction. However my original intention was to obtain a duotone image. And so it was, as you can see in the final photograph at the end of the article.
An (almost) infinite level of detail
Merging so many images leads to a significant increase of the resolution by other means achievable with a single shooting. As already mentioned a set of thirty shots reduced prior to ~ 1.7 Mpx each, returns a single image of about 20 Mpx (cropping included). A useful value for the publication on most of the media without the need of an overpowered computer.
It's interesting to see how we can still go much further ahead. Beyond the most common purposes and stepping into the field of pure experimentation, let's use individual photographs in their original size, 10 Mpx in this case: we get a picture of around 118 Mpx. here's an idea of what you get. From such a large image I cropped some details at 72 dpi resolution (or better known as "100% crops", widespread term, though not precise).
Now it's easy to distinguish individual buildings, from left to right, Citicorp Tower (or Court Square One) - Long Island City, new condominiums in the Rockrose corp. complex - Center Blvd - Long Island City, Met-Life (former Pan-Am) Building -- Midtown, the ruins of the Smallpox Hospital - Roosevelt Island, pylon of the Queensboro Bridge, the Pepsi-Cola billboard along the East River bank since 1936, the Silvercup Studios.
Hours can be spent identifying interesting places.

The final image
As already mentioned, I had in mind a duotonal image as a result. This has influenced exposure and contrast in the capture process.
I prefer the use of traditional techniques for traditional results. However, in this case, having started working digitally, then I carried out the other steps in the same way.
I made an unusual choice for me, in favor of toning: in this specific image it perceptually returns a greater depth. The result does not change much from what could be obtained with a common chemical post printing cycle correctly dosed of bleach and sulfide.
I focused the point of mayor interest in Midtown that seems to emerge upward and just barely downward towards the river, like a bubble blowing up from the horizon line, accompanied by the clouds. CLICK ON THE IMAGE TO ZOOM IN
If you need some clarification or have some questions feel free to ask directly in the comments page (link above).













Read & write comments