8
\$\begingroup\$

I gather that common wisdom says that in addition to camera position you should keep as many variables as possible the same between different shots for a panorama. Variables that spring to mind are (some obviously closely related):

  • Shutter speed
  • Aperture size
  • ISO sensitivity
  • Focus
  • White balance

Which of these is it strictly necessary to lock, and most importantly... why?

For example, in a landscape panorama including the sun, it would be quite difficult (if not impossible) to choose fixed exposure settings for the entire scene. However panorama software I've used typically does a marvelous job of ironing out any differences between images.

Similar questions apply to focus, if you lock focus for a particular shot, can you be sure that others will not be out of focus? (is that likely to happen with a narrow aperture, say f/8?)

\$\endgroup\$
1
  • \$\begingroup\$ For the record I just found a similar related discussion (primarily in the comments) for this older question: photo.stackexchange.com/q/4874/9960 \$\endgroup\$
    – MattJ
    Commented Jun 14, 2012 at 14:23

4 Answers 4

12
\$\begingroup\$

Exposure. Contrary to the conventional wisdom I don't think you should necessarily lock exposure when shooting panoramas. Instead use a metering mode that considers the whole scene (not spot metering), or shoot manual, and get a lot of overlap between images. If you do this the exposure shouldn't vary too much between adjacent shots.

The idea of locking exposure to make images match isn't necessary with current panorama stitching software, which after aligning the images is able to smooth out any changes in brightness.

The problem with locking exposure is that for many scenes there is no single exposure that is correct for the whole range of your panorama, especially if you pan near or past the sun. You have to compromise and accept a level of over exposure on one end and underexposure on the other end. Or you have to shoot multiple exposures for each position and use HDR. Arguably this is the outright best approach but it's a lot more work to good results.

Before I get downvoted it's worth stating that common sense should still be used and if the exposure is jumping around all over the place or the cameras metering isn't doing the job it should be locked.

Focus and aperture should still be locked, however as you can't blend depth of field and blur like you can blend brightness. You state:

Similar questions apply to focus, if you lock focus for a particular shot, can you be sure that others will not be out of focus?

Panoramas aren't usually taken of a single subject, but a whole scene or vista, so something will be in focus in every frame. However using a wide lens and narrow aperture like f/11 and the hyperfocal distance largely mitigates and focus issues.

ISO and shutter speed. In the ideal case you're shooting a static scene using a tripod, in which case set the ISO to the lowest value in order to use a longer shutter speed to capture as much light as you can. Otherwise you might have to chose shutter speed and ISO according to subject / camera motion just as you would if you weren't shooting a panorama. The important thing is you don't absolutely need to lock ISO or shutter speed, small variations wont be a problem, you need want to make sure adjacent images are similar with respect to noise and motion blur.

White balance. Ideally you should be shooting RAW since you are inevitably going to be processing your images. If not, the same advice applies as to exposure.

Focal length should be locked unless you really know what you are doing and are creating a panorama with a particular project in mind.

\$\endgroup\$
6
  • 1
    \$\begingroup\$ I agree. Make sure that the overlap is large and overall scene metering is used. I have had better results with 60% than with 30% overlap. (I.e you have double the amount of images and the middle image is used to get exposure to blend smoothly.) Often though, fixed everything is best as you are treating the panorama as a "single wide lens"-replacement. \$\endgroup\$
    – Unapiedra
    Commented Jun 10, 2012 at 21:46
  • 1
    \$\begingroup\$ Could you show an example using different exposures and how to blend them together? I'm thinking the sky will look unnatural if the exposure isn't kept constant. \$\endgroup\$
    – Eric
    Commented Jun 11, 2012 at 0:48
  • \$\begingroup\$ @Eric Here is one on auto-expose. You can see auto-exposure has significantly darkened the left of the image: matthewwild.co.uk/uploads/haybales.1.min.jpg - however much more exposure on that sky would leave it practically white (I have such a shot in another set where I attempted a more constant exposure, but it's not stitched yet) \$\endgroup\$
    – MattJ
    Commented Jun 11, 2012 at 1:27
  • \$\begingroup\$ Good example! That looks natural enough to me. I'm surprised, because I haven't had that good of results with varying exposures. \$\endgroup\$
    – Eric
    Commented Jun 11, 2012 at 21:33
  • 1
    \$\begingroup\$ @BartArondson if you shoot a single row panorama and stitch it with a rectilinear projection you'll end up with a bow-tie shaped image, cropping to a rectangle will reduce the size of the image. If you were to zoom slightly to shoot the middle section of the panorama then the amount of cropping required could be greatly reduced. This would require the stitching software cope with different focal lengths, and may introduce slight parallax errors for close objects as zooming will likely shift the centre of projection. \$\endgroup\$
    – Matt Grum
    Commented Jun 3, 2013 at 14:45
5
\$\begingroup\$

You'll want to keep the same exposure (shutter, aperture, ISO), focus, and white balance. So all of them.

If the contrast of the entire scene is too much, then you'll need to address that in other ways. Either add artificial lighting to reduce the contrast ratio (e.g. fill flash, reflector) or take additional exposures for HDR.

Why do you want to lock exposure? Because you don't want what should be a continuous tone to look like zebra stripes. :) You'll want to keep the same ISO so that noise is similar. You'll want the same aperture so that DOF is similar. Since you want the same exposure, that locks you into the shutter speed being the same, but that's the least critical component.

Why focus? Having focus change wildly between shots will make for an unnatural-looking panorama.

Same issue with exposure also applies to white balance. You don't want a white wall being more yellow in some areas and more blue in others if all the areas are lit equally.

\$\endgroup\$
2
\$\begingroup\$

Variations on a theme ... :-)

It partially depends on how "arty" or "utilitarian" you are trying to be.
A holiday 360 degree panorama of a mountainscape is probably aimed at getting the scene across.
An architectural view looking at a building with a street running away at an angle either side of centre my be trying to drop out backgrounds or ensure focus all the way up the building tower (or not) or some other effect.

In the latter case a general answer is hard as you need to customise to suit the specific aim.

In the case of a landscape panorama where "conveying the scene" is what matters most then some general comments may end up being relevant (hopefully)

Aperture smaller than larger such that out of focus is less of an issue or not an issue.
Consider manual focus (cf Matt's comment on hyperfocal distance) with focus "touching infinity" and close end of focus as close as that allows. f/22 will pretty much cover everything in most cases (maybe not the daisy at 1 metre, maybe), f/11 pretty much so, f/8 and larger will start to force decisions on what is in focus. Aperture affects this but in most panoramas you will be at relatively low focal length.

  • That said - I shot a 180 degree + 64 frame maybe 100+ kilometre wide panorama of the NZ Southern Alps from a park near the coast in Christchurch. Camera in portrait mode, zoomed in until the mountains to the plains just didn't fill the frame at their visually tallest
    point, start at left end and move in just overlapping frames along whole mountain range. = 60+ images.

Shutter speed is a servant of the situation. As long as there is no dynamic content (animals people cars ...) that you want sharp then as slow as you need after all else is decided. this may be you main "free" variable. Note that clouds and traffic may fool you if not considered.

ISO sensitivity is quality related. Highest ISO that produces quality that suits you completely EXCEPT perhaps where you have a very bright scene and or may be using an ND filter for some effect.

Focus as above for aperture, plus - depends on scene. If there is nothing special or if everything is special you may need to focus frame by frame. If not, then a depth of field that includes the whole scene allows freedom to manually set and forget.

A contrived worst case frame by frame focus case may be standing in the middle of the circle of stones at Stone Henge (should you be allowed)(which you are not) and taking a 360 degree panorama where you wanted the stones in crisp focus but an absolute minimum, depth of field of backdrop beyond. (That would be an awesome photo!). Due to the uneven spacing and sizes and distance to various stones you'd need to focus each frame to suit the content.

White balance I don't recall ever having changed this during a landscape panorama BUT if you were eg shooting towards a sunset towrds the West and evening shadow to the East then a change of WB may well be in order. Auto WB may suit and may not depending on scene and camera.

Exposure - most tricky last - in my experience at least. I've found exposure the parameter I have to be most careful with. To some extent it depends on your stitching software and also on what effect you want. If you want a sunset to deep dusk contrast then changing the exposure to damp down the sunset and bring up the shadows may work exactly against the desired effect. If you have a scene where dynamic range exceeds camera capability (which is often the case and certainly so i the sunset 360 degree example) then a degree of adjustment will be required. This can be just 'autoexposure" but you need to trust your sensors severely. In days of yore (digital but pre-live-view) I would probably have trusted the wide mode light level sensing or done some test shots at key points, set to manual and autoadjusted where required along the way.
With live view (and this will be heresy to some) I am comfortable with using autoexposure per frame, previewing and then adjusting with over-ride (rear thumbwheel) where needed. eg in the example sunset 360 I'd probably drop exposure up to 2 or even 3 stops when shooting towards the sun and maybe raise it one or 2 stops at the extreme dusk portion with progression between these as I go - depending as above on where-ever "even illumination" or "looks like real life" is wanted.

Panorama stitching software makes a difference. I use free and amazingly capable [Autostitch] with a strong tendency towards record-of-journey panoramas rather than anything arty.


On the fly, hand held:

When travelling I will often take informal panoramas of a scene to get an overall feel. this is often done in shortest possible time with hope of absolute perfection a distant dream.
Quick spin around for anything obvious that will cause problems.
Decide what focal length is needed to allow optically largest target frame to look right. (eg a local hill or tree that you want to include will be taller than a distant one .
Set to autoexposure, aperture priority (my default). Aperture as small as reasonable. ISO as high as reasonable. Auto WB usually.
Photograph my hand in closeup :-) (Later unmissable marker of sequence start.

. Frame top left at height to be used as upper bound.
. Half pressure - check liveview exposure level. Adjust if needed. . Either manual focus already set or (usually) point at desired focus point with centre spot focus half pressure, recompose with top left correct.
. Take photo.
. Hold steady - check new target in top right which will be comfortably included in next frame with desired amount of overlap. (This is possibly done as soon as camera is set to new view at start of sequence).

Repeat until finished.

Take photo of my foot in closeup :-) - unmissable sequence end.

The above per-frame procedure may sound cumbersome at first glance but becomes essentially move-check exposure appearance-focus-recompose-shoot-move- ... at maybe a second or two per frame when there is nothing especially special in view.

The above gives me reasonably consistent exposure overall, focust is correct for whatever I want in focus and - two major features - overlap to the extent I have decided I want and a relatively level top line. It is VERY easy to take handheld informal panoramas where the horizon line is an up and down mess so that if you want to cut a rectangle out of it later you lose much material. By formallt siting on something new to set your new horizon and end point to BEFORE you move the frame you get a usefully consistent horizon line. The above is of course not as good as using a tripod and xxx degrees per shot, but works very well for very quick on the fly panoramas.


Revolving restaurant:

Set camera to 30s time lapse.
Place on table pointing out window.
Fiddle with serviette support until view is correct.
Zoom to suit.
East lunch.
Ensure meal is long enough for at least 360 degree sweep.
Ensure frames overlap very significantly (see below).

With enough overlap frames with window bars in may be discarded. Otherwise some cropping needed before stitching.

Set white balance by illuminating white surface with light through window with restaurant light shielded. This allows glass white skew to be removed.

\$\endgroup\$
2
\$\begingroup\$

Everything should be locked. If it can be locked, lock it. If it cannot, try to find a way to make it lock! This includes the position of your lens' nodal point too.

For each deviation from this you reduce the chances of the panorama being stitched correctly. That does not mean you cannot do it, just that your odds of a perfect panorama go down. At the extreme, you can do a hand-held panorama with a point-and-shoot without locking anything. The difficulty of stitching the resulting images depends of subject and lighting conditions. If you choose to shoot an evenly-exposed blank wall, you can get a good stitch no matter what!

Each parameter which varies gives more difficulty to stitching software but you can sometimes correct a poor stitch manually in an image manipulation program like Photoshop. If you understand the basics of how stitching software works, you can get an idea of what is more important to keep constant.

Off the top of my head, this is roughly the order of importance for things which should be kept constant:

  1. Movement in overlap areas. If you have objects like cars of people which cross the areas of overlap and are moving, it can throw off stitching software. If the one you use supports masking, you have to manually mask such areas to remove them from the feature-matching process. Note this includes the shadow of your tripod as the sun or clouds move.
  2. Position of the nodal-point. The closer you have objects in your scene, the more it is important to rotate around the nodal point of the lens. Once objects start moving in relative position to each other or occlusion changes, stitching programs struggle.
  3. Focus. As software is trying to find continuity between images, imagine the difficulty caused by having an object soft in one frame and sharp in another!
  4. Aperture. This one follows from the previous since it affects depth-of-field which would cause objects not match between frames.
  5. Exposure. A good program can read the EXIF and do the adjustment to internally compare features and colors in a wide-dynamic-range space. However, if the exposure changes too much, the right colors and intensities may not be correctly interpolated. If you have a scene of wide dynamic-range it is probably better to bracket and do the panorama in HDR space instead. Sometimes even severe lens vignetting can pose a challenge to stitching software.
  6. ISO & Shutter-Speed are less important because the match is done using a scheme which looks at higher level features first, so changes in noise or slight motion blur have less impact.
  7. White-Balance is probably the least important since you can most times correct before sending images to the stitching program. If the do send the incorrect WB and it differs greatly between images, it is possible the algorithm wont find a match if it tries to inteligently look at colors. Some matching processes work in Luminance space so WB errors cause minimal issues there.

PS: That's what the general idea, some may be missing and the order is not absolute in that it can depend somewhat on the scene being shot.

\$\endgroup\$
5
  • \$\begingroup\$ Any stitching app these days will be using some variation on the Scale Invariant Feature Transform, which can cope with really quite large changes in brightness as it uses gradients rather than absolute pixel values. The key to a good stich is masses of overlap, as that gives many more feature points to work with. \$\endgroup\$
    – Matt Grum
    Commented Jun 11, 2012 at 9:21
  • 1
    \$\begingroup\$ Actually sometimes even high-end software screws up when trying to match features. I often get a better stitch by cropping images and giving it back to the software, particularly when shooting with wide-angle lenses where the perspective distortion makes things hard to match. \$\endgroup\$
    – Itai
    Commented Jun 11, 2012 at 12:34
  • \$\begingroup\$ @Itai So what about in my example of a sunset, that appears full white on what is otherwise the optimal exposure for the rest of the panorama? It blanks out much of the shot, obviously making feature matching much more difficult than if that shot were "correctly" exposed. It just doesn't seem like "always lock everything" is a feasible as a steadfast rule. \$\endgroup\$
    – MattJ
    Commented Jun 11, 2012 at 14:55
  • \$\begingroup\$ @Itai How much overlap are you allowing? In any case SIFT is better at handling brightness variations than it is with perspective transformations (which is pretty much the hardest of all feature matching problems) \$\endgroup\$
    – Matt Grum
    Commented Jun 11, 2012 at 17:35
  • \$\begingroup\$ @Matt - 33% to 25% overlap typically, following the markings on my pano head. Indeed I know how difficult matching is since I worked for 9 years developing image processing, although not stitching related. \$\endgroup\$
    – Itai
    Commented Jun 11, 2012 at 23:36

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.