Is every mosaic a panorama?

No. Panoramas are one type; mosaics can also combine scans, microscope tiles and other image collections.

Why do source images need overlap?

Shared content often provides evidence for alignment and blending, though some calibrated systems can use additional position information.

Can blending fix a misaligned object?

It may hide or soften a seam, but it does not necessarily recover the correct object geometry.

AI Glossary / Computer vision and multimodal AI / Intermediate

Image mosaic

Definition

An image mosaic combines multiple images into a larger or more complete visual representation. In computer vision, the term commonly refers to aligned and composited views of a scene, such as a panorama or stitched scan, rather than a decorative picture assembled from unrelated miniature photographs.

What it means

A camera or scanner may capture only part of the required area at one time. A mosaic brings overlapping captures into a shared coordinate system and combines them. The result can cover a wider field of view or a larger surface than one source image. Its accuracy depends on alignment, the geometric model, exposure consistency and scene changes. A seamless appearance does not guarantee that every measurement is geometrically correct. This matters when mosaics support mapping or inspection rather than only visual presentation.

How it works

A typical pipeline finds overlapping images, estimates correspondences and fits transformations consistent with the capture geometry. It warps images into a common projection, adjusts exposure differences when appropriate and chooses seams or blending weights in overlap regions. Global optimization can reduce accumulated alignment errors across many views.
Different applications need different models. A rotating-camera panorama, a flatbed scan mosaic and an aerial orthomosaic do not share identical geometry. Parallax, moving objects and lens distortion can prevent one simple transformation from aligning everything. Blending can hide some seams but cannot correct arbitrary geometric inconsistency or recover content missing from all source images.

Three practical examples

A photographer stitches overlapping views into a panorama that covers a broader scene than a single frame.
An archive combines overlapping scans of a large poster, checking text and line continuity across joins.
An inspection team assembles tiled microscope images of a flat sample, with calibration suitable for any measurements taken from the mosaic.

Think of it this way

Think of assembling a map from overlapping pieces, aligning shared roads before smoothing the joins. The analogy is limited because photographs involve perspective, lens effects and objects at different depths. Two pieces can disagree locally even when their broad outlines appear to fit.

Worked example

Suppose two ideal planar images are each 1,000 pixels wide and overlap by 200 pixels after alignment, with no scale change or rotation. Their combined horizontal coverage is 1,000+1,000−200=1,800 pixels. The overlap provides common content for registration and blending rather than adding a second copy of that area.
If an alignment error shifts the second image by 5 pixels, a straight line crossing the overlap may appear doubled or broken. Blending can soften the discrepancy without restoring the correct geometry. This simple example omits perspective and projection effects but shows why coverage and alignment accuracy are separate concerns.

Illustrative case study

Fictional example: Cedar Archive scans a large diagram in four overlapping sections. Automatic stitching aligns most text, but a fold causes local distortion near one join. The team checks the original, revises the alignment and marks an area where exact geometry remains uncertain. It preserves the component scans and transformation settings. The case illustrates that a visually complete mosaic can still require local review, especially when the source surface is not perfectly flat.

Common misconceptions and mistakes

An image mosaic is not necessarily an orthorectified map. A panorama projection can distort angles and lengths while looking natural to a viewer. More overlap can help matching but does not guarantee correct correspondences. Exposure blending does not solve parallax. Moving objects may be duplicated, cut or removed by seam choices. Finally, the output pixel count is not a direct measure of independent detail, because overlap, resampling and blur affect the information actually retained.

Business use

Businesses should define whether the mosaic is for presentation, search or measurement. Choose capture geometry and calibration accordingly. Record source coverage, transformations, projection and blending settings, and retain source images for audit or reprocessing. Inspect seams around text, edges and moving objects. For measurement applications, validate positional accuracy against references rather than approving only a smooth appearance. A production workflow should detect failed alignments and incomplete coverage instead of silently exporting an apparently finished but unreliable composite.

Check your understanding

Two aligned 800-pixel-wide images overlap by 100 pixels with no scaling. What is their combined width?

Answer: 1,500 pixels. The shared 100-pixel region is counted once, assuming the simplified planar layout and no additional cropping.

Related terms

Image stitching
Panorama
Image registration
Bundle adjustment
homography
seam finding
orthomosaic

Learn next

Learn image overlap and registration, then stitch a small planar example and inspect its seams. Study projection models and parallax before using mosaics for larger scenes or physical measurements.

Sources

OpenCV, High level stitching API: https://docs.opencv.org/4.10.0/d8/d19/tutorial_stitcher.html OpenCV, image-stitching tutorials: https://docs.opencv.org/4.1.2/d0/d33/tutorial_table_of_content_stitching.html Official implementation excerpts consulted; application-specific geometric and editorial review remain pending. Coverage arithmetic and the fictional case are original.

Previous: Image morphing Next: Image noise