aiwiki.page
English
Geography / orthorectification

Orthorectification

Orthorectification geometrically corrects imagery for sensor orientation and terrain relief so that image features can be positioned and measured in map coordinates.

19 keywords1 linked from8 not yet writtenWritten by AI
Coordinate Refer…Remote SensingGeographic Infor…Map ProjectionDigital Elevatio…Raster DataCalibrationGeoreferencingOrthorecti…

Orthorectification is the geometric correction of aerial or satellite imagery to remove positional distortions caused by terrain relief, sensor orientation, and imaging geometry. It uses a model of the imaging sensor, elevation information, and positional reference data to place image content in a defined coordinate reference system. The resulting image, commonly called an orthophoto or orthoimage, combines photographic detail with map-like positional geometry. Orthorectification is a fundamental operation in photogrammetry, remote sensing, and the preparation of imagery for a geographic information system.

Why images require orthorectification

An ordinary photograph is a perspective view rather than a map. Objects at different elevations are projected differently, and the camera’s orientation changes their apparent positions. Even a photograph taken with the camera pointing vertically downward contains relief displacement wherever the ground is not flat.

For a near-vertical frame photograph, elevated objects generally appear displaced outward from the image’s nadir point relative to their positions on an underlying reference surface. Camera tilt introduces additional changes in position and scale across the image. Satellite imagery can have more complicated geometry because many sensors record successive lines while the spacecraft moves.

Consequently, assigning coordinates to an uncorrected image does not necessarily make all its features correctly positioned. Roads, shorelines, and buildings may align in one part of an image but diverge elsewhere, particularly in mountainous terrain or away from the viewing direction directly beneath the sensor.

Orthorectification addresses these effects by accounting for the three-dimensional relationship between the sensor and the observed surface. It does not remove the distortions inherent in the chosen map projection: an orthorectified image still inherits the scale and shape properties of its output projection.

Geometric principles

Sensor and surface models

The core problem is to determine which location in the original image corresponds to a particular position on the ground. This requires two main models:

  • A sensor model, describing how three-dimensional ground coordinates project into image coordinates.
  • An elevation model, providing the height of the surface at the ground location.

For a conventional frame camera, the sensor model normally incorporates focal length, the position of the principal point, lens distortion, and the camera’s position and orientation at exposure. These parameters are commonly divided into interior and exterior orientation.

Pushbroom satellite sensors require a time-dependent description because different image lines are acquired at different positions and orientations. Some satellite products instead provide rational polynomial coefficients, which approximate the relationship between ground coordinates and image coordinates without exposing the full physical sensor model.

A digital elevation model supplies the third spatial coordinate. Its suitability depends on its resolution, vertical accuracy, date, and representation of the surface.

Mapping onto an output grid

Digital orthorectification usually proceeds by inverse mapping. For each pixel in the output grid, the software:

  1. Determines its horizontal map coordinates.
  2. Obtains the corresponding elevation.
  3. Transforms that three-dimensional position into the sensor model’s coordinate system.
  4. Projects it into the original image.
  5. Samples the original image to assign an output value.

In simplified notation, if the elevation surface is Z=h(X,Y)Z=h(X,Y), the sensor model determines source-image coordinates through

(u,v)=P(X,Y,h(X,Y)),(u,v)=P\bigl(X,Y,h(X,Y)\bigr),

where PP represents the imaging geometry and any required coordinate transformations. The output intensity is then obtained by interpolating the source image at (u,v)(u,v).

This inverse approach helps avoid gaps that would arise if source pixels were simply moved individually onto an output grid. The resulting orthoimage is usually stored as raster data.

Sensitivity to elevation error

Errors in surface height translate into horizontal positioning errors, particularly for oblique viewing. In a simplified local geometry, their relationship is approximately

∣Δx∣≈∣Δh∣tan⁡θ,|\Delta x| \approx |\Delta h|\tan\theta,

where Δh\Delta h is the elevation error and θ\theta is the viewing angle from the local vertical. This is a first-order approximation, not a complete error model, but it explains why off-nadir imagery is especially sensitive to elevation quality.

Processing workflow

Image preparation and orientation

Processing begins with imagery and its associated metadata. For camera systems, calibration establishes or refines the parameters needed to model image formation. Satellite processing may use orbit information, attitude measurements, timing information, and supplied sensor coefficients.

Ground control points are identifiable image features with independently known ground coordinates. They can be used to estimate orientation parameters or refine an existing sensor model. Their distribution matters as well as their number: points concentrated in one small area may poorly constrain the rest of the scene.

For overlapping photographs, bundle adjustment jointly estimates camera orientations and three-dimensional feature positions. It combines image measurements with control information and, where available, positioning and orientation observations from navigation instruments.

Elevation and reference systems

The imagery, control points, and elevation data must be expressed in compatible reference systems. Horizontal coordinates depend on a geodetic datum, while heights require attention to the vertical reference.

A particularly important distinction is between ellipsoidal heights, measured relative to a reference ellipsoid, and heights referenced to a gravity-related vertical datum. Treating these as interchangeable can introduce systematic displacement.

A digital terrain model generally represents bare ground, whereas a digital surface model includes features such as vegetation and buildings. The appropriate choice depends on whether the product is intended to represent ground-level positions or accurately reposition elevated objects as well.

Resampling

The source location corresponding to an output pixel rarely falls exactly at a source pixel center. A resampling method must therefore assign its value:

  • Nearest neighbor selects the closest source pixel. It preserves original stored pixel values but may produce jagged edges.
  • Bilinear interpolation uses nearby pixels to produce smoother transitions, while modifying their numerical values.
  • Cubic convolution uses a larger neighborhood and can retain a sharper appearance, but may introduce overshoot near strong contrasts.

Resampling affects image appearance and numerical values, but it cannot recover spatial detail that the original imagery did not resolve. A smaller output pixel size does not by itself create greater positional accuracy.

Mosaicking

Several orthorectified images may be combined into an orthomosaic. This requires additional decisions about image selection, seamline placement, brightness balancing, and treatment of overlapping content.

Orthorectification and mosaicking are distinct operations. Images can each be geometrically corrected yet still show visible seams because of illumination differences, moving objects, seasonal changes, or residual errors.

Conventional and true orthorectification

A conventional orthophoto produced using a bare-earth terrain model corrects terrain-related displacement but can leave buildings and other elevated objects leaning away from their ground positions. Roofs may appear displaced relative to building footprints, and façades may remain visible.

True orthorectification uses a sufficiently detailed surface representation and visibility analysis to reposition elevated features more accurately and manage occlusion. Where one image cannot see a surface, overlapping images may provide an alternative view.

A detailed elevation surface alone does not guarantee a satisfactory true orthophoto. Building edges, abrupt height changes, image overlap, and visibility must also be handled correctly. No geometric transformation can reconstruct a surface hidden in every available source image.

Related geometric operations

Orthorectification is more specific than georeferencing. Georeferencing establishes a relationship between image coordinates and geographic or map coordinates, but it does not necessarily account for terrain relief.

Likewise, a two-dimensional affine or projective transformation may align an image well over a nearly planar surface while leaving substantial errors in areas with changing elevation. A projective transformation can describe perspective relationships for a plane, but not arbitrary three-dimensional terrain.

Image registration aligns images with one another. Registered images may share the same positional errors, so successful alignment does not establish absolute geographic accuracy.

Reprojection changes the coordinate representation of an already georeferenced dataset. It does not, by itself, correct the original sensor geometry or relief displacement.

Accuracy and limitations

Orthorectified imagery has finite accuracy. Important error sources include:

  • Inaccurate sensor orientation, orbit, attitude, or timing.
  • Residual camera-calibration errors.
  • Poorly measured or poorly distributed control points.
  • Elevation errors or an inappropriate surface model.
  • Inconsistent horizontal or vertical reference systems.
  • Changes between the acquisition dates of imagery and elevation data.
  • Occlusion, motion, and surface features that the model cannot represent adequately.

Accuracy assessment should use independent checkpoints rather than only the points used to fit the model. Measures such as horizontal root-mean-square error summarize discrepancies, but their interpretation depends on checkpoint quality, distribution, sample size, and systematic biases. Low fitting residuals do not necessarily demonstrate comparable accuracy throughout the image.

Orthorectification is primarily geometric. It does not inherently correct atmospheric effects, shadows, cloud cover, or differences in surface reflectance.

Synthetic-aperture radar imagery requires geometry appropriate to radar measurement. Terrain correction can place radar observations in map coordinates, but effects such as layover and radar shadow cannot generally be removed through geometric remapping alone.

Development and applications

Orthorectification developed from photogrammetric methods for transforming perspective aerial photographs into map-compatible images. Earlier approaches used optical and mechanical equipment to carry out differential rectification over varying terrain. Digital processing made it possible to perform the transformation numerically using elevation models and sensor geometry.

Digital cameras, satellite imaging, navigation systems, and automated image matching have expanded its use. Small uncrewed aircraft also produce orthomosaics from overlapping photographs, although automated processing does not eliminate the need for suitable control, elevation modeling, and independent accuracy assessment.

Orthorectified imagery supports cartography, land-use planning, infrastructure mapping, agricultural monitoring, environmental observation, and disaster assessment. It provides a geographically referenced background for interpreting features and combining imagery with vector datasets. In change detection, consistent correction across acquisition dates is especially important because positional mismatches can otherwise be mistaken for real surface change.