Mosaic vs. Panorama
Both mosaic and panoramas are made up of multiple pictures stitched together to form a larger image. You’re probably familiar with panoramic pictures; you keep the camera in one place and take pictures in different directions. Mosaics (or at least the way I’m defining them) keep the camera pointed in the same direction, but physically move it to different locations. Panoramas are typically done with larger objects (such as a landscape) because it’s easier to change the direction of the camera than it is to move it horizontally (potentially miles). However, when used with smaller objects or things that aren’t perfectly flat, panoramas will look distorted. Mosaics preserve the shape better.
What makes a good Mosaic?
To make the stitching-together process easier and introduce fewer artifacts, the object should be more two dimensional than three. Having a lot of textures and details are useful (and adds to the beauty), but being too shiny will not work unless the reflections are carefully controlled. Musical instruments or assemblages are perfect examples.
Setting up the subject
As with all photography, lighting is key. You’ll want a wide, even light source, such as a big window, reflections off a white wall, or multiple diffuse lights. You don’t want a point light source (like a flash) whose bounce will create hotspots on the object. Worst is an on-camera flash; this will create a distracting repetitive pattern that highlights each individual picture. The goal is to keep the light to be even no matter the position of the camera.
A typical studio setup has even light from both sides, plus some from behind to highlight the edges. In my setup, I used reflections off a white wall to provide the backlight. Be conscious of your background to make sure it isn’t too distracting.
Notice the relative lack of shadows on the wall due to the large reflectors.
Finally, since macro photography shows any dust on the subject, you’ll probably want to wipe it down before shooting — but, of course, if the dust adds character, leave it in! A lot of the fuzziness in the Delsonics guitar above is actually heavy dust.
Controlling Parallax / Planning the shots
Parallax is the single biggest non-obvious problem you’ll encounter, but with careful planning it can be minimized. Parallax is the apparent movement of closer objects more than further objects (the wikipedia article has some examples), and it is caused by moving the camera between shots. The problem and the cause are best shown by example:
These strings are an extreme example of the problems parallax can cause. I’ve taken two pictures, one to the left of the other. The stitching program has aligned the images so that the fingerboard inlays and the shadow of the string align, but the strings, being closer to the camera than the fingerboard, are impossible to simultaneously align.
I’ve solved the problem by taking the two pictures - one above the other, in the direction of the string. The dividing line is still obvious (mainly because my lighting wasn’t even enough), but this would be much easier to retouch than the previous example.
So, to minimize parallax, you’ll need to plan your shots so the camera follows the non-2-dimensional objects. This means that, for a stringed instrument, you can’t move the camera horizontally, so you’re limited to only a column of pictures in the direction of the strings.
I did some test shots with my electric cello so I could perfect my technique and not waste my friend's studio time. I’ve divided the mosaic in to 7 shots, making sure I have ample overlap so that the computer can better align the images:
The slight left/right movement of each picture is for illustration only; in reality they were all aimed at the dead center of the instrument to minimize parallax along the strings.
For the more complicated non-2-dimensional parts of the cello, I controlled parallax another way: by including the feature in only one photo, so that there is no ambiguity when it is merged. Examples of this are the pegbox in the top picture, the bottom of the fingerboard, and the bridge in the bottom picture.
Setting up the camera
For the camera, pick a medium telephoto lens, somewhere in the 70-200mm 35mm equivalent format. This minimizes parallax distortions that would appear when wider-angle lens images are stitched together.
Set the aperture to a mid range so that you get enough depth of field for your subject. Don’t close it all the way down, or the background will be in focus and will become more distracting.
Find a good exposure and then keep the camera in manual mode. This ensures that all the photos are at the same exposure and helps keep the stitching look seamless. You can also use this opportunity to make sure the object is lit evenly, which is particularly useful if using multiple flashes.
Place the camera so that the object mostly fills the frame in one direction (in my case, it was the horizontal) - this means the extra pixels are not wasted and you get maximum resolution. But, also, when doing this, make sure you’re not getting closer than the minimum focus distance.
Taking the photographs
Once the lighting, object, and camera are set up, taking the planned sequence of pictures is fairly quick.
A tripod will be helpful to ensure you are able to shoot the object from the exact same distance and direction every time. I raised and lowered the tripod between pictures, and was careful to make sure the center pole was over the exact same spot in the rug every time.
Don’t rush it - make sure that there is plenty of overlap. And don’t strike the setup as soon as you’re done - you’ll learn things when you try and stitch the photos together, so plan on a couple of attempts before getting it perfect.
There are a lot of different software packages that will allow you to stitch together mosaics. I’m going to use Hugin, an open source program that, while not always the easiest-to-use, is very powerful.
I’m not going to cover how to use Hugin in detail, but instead concentrate on how to use it for mosaics instead of regular panoramas. It comes with a good set of tutorials, of which, I found these two most helpful:
Here's the general workflow I used:
* Under "Interface" on the menu bar, select the “Advanced Interface”. This is important because without it, you won’t be able to find the selections below.
* Under the “Photos” tab, click “Add images” to load your files.
- Make sure each photo is listed as a different lens. This is a little weird, but it tells the program that the lens moved between shots. (highlighted on figure below)
- Try to automatically align the pictures by using the “Create control points” button. If this fails, you’ll have to do it manually using the “Control Points” tab. Here you show the program how the pictures match each other by picking points correspond between the multiple pictures. This is the most time-consuming part but also the place were you can make the most improvements to the final quality.
* Under the "Photos" Tab, "Optimize" section, select “Geometric: Custom Parameters” (see figure below). This will make a new tab appear, named “Optimizer”
* Go to the new “Optimizer” tab. The computer needs to put together a model of how the photos fit - the locations and tilts of each picture relative to each other. Since we’ve been careful to take each photo in exactly the same direction, we can tell the computer to assume that they are all parallel and to stitch it together assuming only that the camera has moved. If we don’t do this, it will make guesses that will probably be wrong.
- The first thing is to pick a reference image - one that whose alignment is “perfect” and all the other ones will be transformed to match it. I pick the center-most image and let it build the image outward from there. For this image, set all parameters to zero (click on the parameter and type ‘0’). Then make sure they are fixed (option-click on the parameter until it is not underlined)
- Next, we tell the computer what adjustments are permissible for the non-reference images. Since we kept the camera pointed in the same direction, we’ll keep the Yaw and Pitch set to zero. To do this, type click each parameter, type zero, and make sure the parameter is not underlined. We will allow it to adjust the Roll, X, Y, and Z (option-click to make the parameters underlined, and still enter 0 as a starting point). Theoretically roll adjustments (turning the camera to fix an uneven horizon) should not be needed, but the computer generally won’t make bad guesses for this parameter. The XYZ is the position of the camera; the computer will need to solve for this to make the mosaic work.
- Click the “Optimize now!” button to computer the orientations of the photos. You can preview this in the “View > Preview Window” menu item.
Then to finish it all up:
* Go to the “Stitcher” tab to make the final image. Set the Projection to "Rectilinear". Click “Calculate field of view”, “Calculate optimal size”, and “Fit crop to images”. Save your project and click “Stitch!” to make the magic happen.
The Finished Image (draft)
Ok, ok, what does a super-detailed mosaic look like? Here's my best raw, unedited test-run I did with my cello. This is one-seventh (15%) size; click for half-size:[Clipart: openclipart.com ]