A flexible and high-accuracy framework for modeling,
applying and inverting optical distortion.
Unlike conventional calibration libraries, Optical Distortion Lab
is built around a forward-first distortion philosophy.
Distortion is explicitly defined, analyzed, and validated
before inverse mapping is applied.
Unlimited Polynomial Models
No predefined limit on polynomial degree or number of terms.
Models are constrained only by mathematics and physical validity.
Monotonicity & Physical Validity
Automatic detection of non-monotonic mappings, folding,
and irreversible information loss.
Robust Inverse Mapping
LUT-based inverse mapping with stable interpolation.
Production Ready
Script-based batch processing, distributed as .mltbx.
Custom Sampling Control
Balance accuracy, performance and output resolution.
Visual Evidence: Monotonic vs Non-Monotonic Distortion
Monotonic Distortion Model
Distorted
Undistorted
Monotonic
Radial Mapping
Non-Monotonic Distortion Model
Distorted
Undistorted
Non-Monotonic
Radial Mapping
Technical note:
The two distortion models differ by a single higher-order term
of -0.05 · r11.
This term has negligible influence near the image center,
but dominates toward the outer field, causing a transition
from monotonic to non-monotonic radial mapping.
What This Toolbox Is Not
No Parameter Estimation (Yet)
The toolbox does not currently estimate distortion parameters
from images or calibration patterns.
Not a Black Box
All distortion models are explicitly defined and inspectable.
There are no hidden assumptions or locked equations.
This is a deliberate design choice.
Reliable estimation requires a well-defined and physically valid
forward model — not the other way around.
The focus is on forward modeling and physical correctness,
which form the foundation for any future estimation capability.
Why Not OpenCV or Standard Calibration Tools?
OpenCV and similar libraries are excellent for
parameter estimation and camera calibration.
However, they are fundamentally designed around
inverse correction, not forward distortion modeling.
Low-Order Model Assumptions
Most calibration tools rely on predefined,
low-order distortion models that implicitly assume
smooth and monotonic behavior.
No Physical Validity Check
Non-monotonic mappings, folding, and information loss
are usually not detected or reported explicitly.
Inverse-Only Perspective
The forward distortion function is rarely treated
as a first-class, inspectable object.
Optical Distortion Lab addresses these limitations by
treating forward distortion as the primary problem,
enforcing physical constraints, and allowing
unrestricted model complexity when required.
Roadmap & Future Work
Distortion Estimation
Parameter estimation modules built on top of
the existing forward distortion engine.
Python Implementation
A Python version targeting machine learning
and large-scale dataset generation.
Differential Calibration
Physically constrained differential calibration
methods for advanced optical systems.
Licensing & Availability
Optical Distortion Lab is distributed as a
commercial MATLAB toolbox,
intended for research, development, and industrial use.
Research & Academic License
Suitable for universities, research groups,
and individual researchers.
Commercial License
For industrial applications, product development,
and internal tooling.
Evaluation Access
Limited evaluation versions may be provided
upon request.
