AI Glossary / Computer vision and multimodal AI / Intermediate

Brown - Conrady model

Definition

The Brown - Conrady model is a camera-distortion model that describes departures from ideal perspective projection using radial and tangential terms. It is widely used in calibration to estimate how image coordinates are displaced by lens distortion and to support geometric correction.

What it means

An ideal pinhole camera maps straight scene lines to straight image lines, apart from projection effects. Real lenses can bend those lines or shift image points in systematic ways. Radial distortion varies with distance from the distortion center, while tangential or decentring terms represent certain asymmetric effects.
The model uses a small set of coefficients rather than a separate correction for every pixel. Those coefficients are estimated from calibration observations, together with relevant camera parameters.
Different implementations use different extensions, coefficient orders and forward or inverse conventions. A coefficient value or sign is therefore meaningful only alongside the exact equations and coordinate system in which it is used.

How it works

A common forward form starts with undistorted normalized coordinates x,y and r²=x²+y². Radial scaling uses a polynomial such as 1+k1r²+k2r⁴+k3r⁶. Tangential terms involving p1 and p2 add coordinate-dependent offsets. Intrinsic camera parameters then map normalized coordinates to pixels.
Calibration fits the parameters to observed image locations of known or constrained scene points. Undistortion uses the model to remap the image or correct point coordinates, often requiring numerical inversion of the forward relationship. Check the range where the fitted model is reliable. Wide-angle or unusual optics may require a different model rather than simply adding coefficients without validation.

Three practical examples

• A measurement system corrects lens-induced curvature before estimating dimensions from calibrated images.
• A stereo setup uses calibrated distortion parameters before rectifying the two camera views.
• A reconstruction pipeline accounts for lens distortion when projecting three-dimensional points into photographs.

Think of it this way

Think of a regular grid drawn on an elastic sheet that has been stretched unevenly around its center. Calibration estimates a compact description of that deformation. The analogy explains geometric displacement, but lens distortion is an optical mapping and correction requires the right coordinate convention; it does not undo every blur or optical imperfection.

Worked example

Use a simplified forward radial model with k1=−0.1, all other distortion coefficients zero, and normalized point (x,y)=(0.5,0.4). Then r²=0.25+0.16=0.41. The radial scale is 1−0.1×0.41=0.959.
The distorted normalized coordinates become xd=0.5×0.959=0.4795 and yd=0.4×0.959=0.3836. Under this specific forward convention, the point moves toward the center. Pixel coordinates would require the camera's focal-length and principal-point parameters.
This example is not an undistortion formula: applying the same forward equation again would distort the point further rather than necessarily recover its original position. It also omits tangential terms for clarity. Real calibration must use enough well-distributed observations to estimate the chosen model and assess residual errors across the image.

Illustrative case study

Fictional case: Ash Measurement transfers calibration coefficients from one vision library to another. The second implementation expects a different coefficient ordering, and corrected images become more distorted near the edges.
The team checks the equations, coordinate units and mapping direction, then validates the conversion on known calibration points. It also confirms that the camera's focus and image-processing configuration match the calibration conditions. The case illustrates implementation discipline. It does not establish that any fixed coefficient set remains valid after changing the lens, crop, resolution or other relevant imaging conditions.

Common misconceptions and mistakes

Brown - Conrady is not only a generic barrel-distortion slider; common forms include tangential terms and specific coordinate conventions. Coefficient signs should not be interpreted without knowing the mapping direction.
Geometric undistortion does not recover detail lost to blur, noise or saturation. A low average calibration error can conceal poor edge coverage or systematic residuals. Reusing coefficients across changed imaging configurations requires care. More polynomial terms do not automatically improve accuracy and may produce unstable behavior outside the observed calibration region.

Business use

The model is important when image coordinates support measurement, stereo depth, mapping or accurate overlays. For casual display, small distortion may be acceptable; for geometric analysis it can become a systematic source of error.
Document the camera, lens, focus, image size, coefficient order and equation convention. Retain calibration images and evaluate residuals throughout the field of view. Validate correction on observations not used for fitting where practical. If the lens is unusually wide-angle or the residual pattern remains structured, compare suitable alternative camera models. A calibration file should be treated as a measured configuration with limits, not a universal camera identity.

People also ask

Does it model every lens perfectly?

No. It is an approximation whose suitability must be checked.

Are coefficients expressed in raw pixels?

Common formulations use normalized coordinates, but implementations must be checked.

Is distortion correction the same as sharpening?

No. It remaps geometry; sharpening addresses a different aspect of image appearance.

Check your understanding

A forward distortion model maps an ideal point to its distorted location. Can undistortion always be performed by applying that identical formula again?

Answer: No. Undistortion requires the inverse mapping, which may be solved numerically or implemented through a suitable remapping procedure. Mapping direction must be explicit.

Related terms

• Camera calibration
• radial distortion
• tangential distortion
• camera intrinsics
• undistortion
• pinhole camera model
• stereo rectification
• reprojection error

Learn next

Learn normalized camera coordinates and intrinsic parameters next. Trace one point through ideal projection and forward distortion, then examine how a library performs the inverse correction. Compare residuals across the image before accepting a calibration model.

More in Computer vision and multimodal AI

• Bounding box - A bounding box is a rectangular region used to indicate the location and approximate extent of an object or other area of interest in an image.
• Boustrophedon cell decomposition - Boustrophedon cell decomposition divides a robot's free workspace into regions that can be covered with orderly back-and-forth passes.
• Box filter - A box filter is an image filter that combines values within a rectangular neighborhood using equal weights.
• Bundle adjustment - Bundle adjustment is an optimization procedure that jointly refines camera parameters and, in many reconstruction settings, three-dimensional point...
• Burst photography - Burst photography captures a sequence of images in rapid succession.
• Burst super-resolution - Burst super-resolution reconstructs a higher-resolution image from multiple rapidly captured frames that contain complementary spatial samples.

Sources

OpenCV, Camera Calibration tutorial: https://docs.opencv.org/doc/doxygen/html/dc/dbb/tutorial_py_calibration.html OpenCV, Camera Calibration reference: https://docs.opencv.org/5.0/main_modules/calib.html Official documentation excerpts consulted. The numerical example states a simplified forward convention; full geometric and editorial review remain pending.

Previous: Brown clustering Next: Brown - Forsythe test