How Photo Stitching Works: From Multiple Images to One Panorama

Step 2: Feature matching

Next the software compares descriptors between photos to find the same spot in two images. Many candidate matches are wrong, so it filters them. One common filter keeps a match only if it is clearly better than the second-best candidate. Another uses a geometric test that checks whether most matches agree on the same movement between the two images, and discards the ones that do not. Repeated patterns, such as rows of identical windows, are classic sources of false matches.

Overlap makes matching possible. That is why a panorama guide tells you to overlap photos by about a third: the overlap is where matches come from.

Step 3: Alignment

From the matches, the software estimates how the camera turned between photos. For panoramas taken by turning on the spot, the motion is mostly a rotation, which makes the maths well behaved. All photos are then adjusted together so that matched details agree as well as possible across the whole panorama, a process related to bundle adjustment.

Panorama 360 has an advantage here: the phone's motion sensors report which way the camera was pointing for each photo. That gives the stitcher a good first estimate for every image, so matching only needs to correct small errors. If a scene has little detail to match, the sensor estimate does most of the work, and the app tells you so.

Step 4: Projection

All photos now need to be drawn on one surface. For a flat wide panorama, a cylinder is common. For a full sphere, the result is stored as an equirectangular image where horizontal position is longitude and vertical position is latitude. The spherical 360° panorama page explains that format. Projection also determines the file's shape: a full sphere is a 2:1 image, while a horizontal 360° is a very wide, short strip.

Step 5: Exposure compensation

Cameras adjust exposure automatically, so photos taken moments apart can differ in brightness, especially when you turn from a bright window to a dark wall. The stitcher estimates a gain for each photo so that overlapping areas match. Locking exposure during capture reduces the problem before it starts. Panorama 360 locks exposure after the first photo and also balances brightness during stitching.

Step 6: Blending

Finally, the overlaps are blended. A hard cut would show a visible line, so the software fades from one photo to the next across the overlap. Simple feathering uses a gradual weighting from the edge of each photo. More advanced multi-band blending handles different levels of detail separately. Blending hides small brightness and alignment differences. It cannot hide a large misalignment or something that moved.

Parallax, ghosting and seams

Three problems account for most visible errors:

• Parallax. If the camera moves sideways while you turn, near objects shift relative to far ones, and no single alignment is right for both. See parallax on Wikipedia. Rotating on the spot avoids it, and professional panorama heads rotate around the lens's entrance pupil for the same reason.
• Ghosting. A person or car that moves between photos appears twice, partly, or cut in half in the overlap.
• Seams. Visible lines appear where alignment or brightness is imperfect.

Why some scenes are difficult

• Low-texture scenes. Blank walls, clear sky and calm water offer few features.
• Repeating patterns. Fences, tiles and rows of windows confuse matching.
• Strong close-ups. Near objects cause parallax.
• Moving subjects. Crowds, traffic and foliage in wind change between photos.
• Extreme contrast. Bright windows and dark interiors strain exposure.
• Blur. Soft photos lose features.

What this means for you

Most stitching quality is decided at capture. Keep the phone level, rotate on the spot, hold still, overlap, avoid movement and keep your distance from close objects. Our guide to 12 practical panoramic photography tips turns that into a checklist, and 10 common panorama photography mistakes helps you diagnose problems from the symptoms.

Stitching on a phone versus on a computer

Desktop software can use a lot of memory and take its time. A phone has less of both, so a phone stitcher has to be economical. A common approach is to do the heavy alignment work on reduced-size copies of the photos and then render the final image at full detail in pieces, so that it never holds everything in memory at once. That is why phone results are limited in size on low-memory devices, and why a progress display and a cancel button matter. Panorama 360 works in this way and reduces the output size on phones with little memory, instead of crashing.

Stitching in Panorama 360

Panorama 360 stitches on your phone. It starts from the sensor orientation of each photo, matches details between neighbours to refine the alignment, balances exposure and blends the overlaps, then classifies the result as a full sphere, a horizontal 360° panorama or a partial panorama. If there are too few photos or a photo cannot be read, it reports an error rather than returning something misleading. See the panorama photo stitcher page for the app's workflow. Panorama 360 is coming soon to Android and is not on Google Play yet.