Orthorectification and Affine Transformation

Orthorectification and affine transformation are two common geometric transformation methods in geospatial data processing. They differ significantly in application and purpose.

Orthorectification

Orthorectification is the process of correcting remote sensing imagery or aerial photographs that contain terrain relief and sensor geometric distortion into a nadir ground view, such as a map, so that the position of each pixel in the image strictly corresponds to its real-world ground coordinates. The method starts from a camera or sensor model, such as RPC or a polynomial model, combines it with a digital elevation model (DEM), projects each image pixel into three-dimensional ground space, and then resamples it onto a unified projection plane.

In short:

It straightens a tilted image and corrects for uneven terrain.

Characteristic	Description
Correction target	Terrain relief, tilt, and sensor imaging distortion
Input data	Raw image + digital elevation model (DEM) + RPC or geometric model
Output image	Orthophoto / orthoimage
Accuracy	Highest; suitable for demanding applications such as measurement, cartography, and GIS analysis
Overlay capability	Can overlay seamlessly with maps and GIS layers without geometric error
Common uses	Aerial photogrammetry, satellite remote sensing, and UAV mapping

---

Affine Transformation

An affine transformation is a two-dimensional geometric transformation that preserves straight lines and parallelism. It is used for image geometric correction, such as image registration, mosaicking, and projection transformation. It uses linear algebra to implement an approximate linear mapping between image coordinates and ground coordinates.

Affine Transformation Parameters in GDAL

GDAL uses an array of length 6, called GeoTransform, to describe the affine relationship from pixel space, or raster indices, to geographic space, or map coordinates:

GeoTransform[6] = [GT0, GT1, GT2, GT3, GT4, GT5]

This transformation defines the mapping from raster row and column (i, j) to geographic coordinates (X_geo, Y_geo):

$$X_{geo} = GT0 + GT1 * i + GT2 * j$$ $$Y_{geo} = GT3 + GT4 * i + GT5 * j$$

Parameter meanings, based on the official GDAL definition:

Parameter index	Name	English description	Meaning
GT0	TopLeftX	X coordinate of the top-left corner of the top-left pixel	Map X coordinate of the upper-left pixel, usually projected or geographic coordinates
GT1	PixelWidth	W-E pixel resolution / pixel size in X direction	Size of each pixel in the X direction, usually positive
GT2	RotationX	Row rotation (typically 0)	X component affected by row-direction rotation, usually 0
GT3	TopLeftY	Y coordinate of the top-left corner of the top-left pixel	Map Y coordinate of the upper-left pixel
GT4	RotationY	Column rotation (typically 0)	Y component affected by column-direction rotation, usually 0
GT5	PixelHeight	N-S pixel resolution / pixel size in Y direction (usually negative)	Size of each pixel in the Y direction, usually negative to indicate downward image rows

References:

Raster Data Model - GDAL documentation

Affine Transformation in GeoTIFF

Affine transformation is the core mechanism that accurately connects GeoTIFF imagery to real-world geographic coordinate systems. GDAL, and most GIS software, uses six GeoTransform parameters plus projection information, such as WKT or EPSG, to transform an image from pixel-grid space into geographic space.

How affine transformation is stored in GeoTIFF:

• TIFF tags:
  • ModelPixelScaleTag -> corresponds to GT1 and GT5, the pixel size
  • ModelTiepointTag -> corresponds to GT0 and GT3, the image reference point
• GeoKeyDirectoryTag:
  • Stores CRS information such as EPSG code, units, and projection type
• GDAL metadata, in an external .aux or VRT file:
  • GeoTransform explicitly represents all six parameters

Comparison Summary

Item	Orthorectification	Affine Transformation
Accounts for elevation	Yes; accounts for terrain relief with a DEM	No; ignores terrain height differences
Geometric model accuracy	High; supports sensor modeling	Medium; linear model
Required data	Image + DEM + geometric model/RPC	Image + control points
Application scenarios	Aerial orthophotos and satellite image mapping	Image registration and minor terrain correction
Transformation property	Nonlinear projection transformation	Linear geometric transformation
Accuracy and cost	High accuracy but computationally complex	Medium accuracy but fast