Skip to content

📷 Lens distortion correction for Python, a wrapper for lensfun

License

Notifications You must be signed in to change notification settings

letmaik/lensfunpy

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

lensfunpy

lensfunpy is an easy-to-use Python wrapper for the lensfun library.

API Documentation

Sample code

How to find cameras and lenses

import lensfunpy

cam_maker = 'NIKON CORPORATION'
cam_model = 'NIKON D3S'
lens_maker = 'Nikon'
lens_model = 'Nikkor 28mm f/2.8D AF'

db = lensfunpy.Database()
cam = db.find_cameras(cam_maker, cam_model)[0]
lens = db.find_lenses(cam, lens_maker, lens_model)[0]

print(cam)
# Camera(Maker: NIKON CORPORATION; Model: NIKON D3S; Variant: ;
#        Mount: Nikon F AF; Crop Factor: 1.0; Score: 0)

print(lens)
# Lens(Maker: Nikon; Model: Nikkor 28mm f/2.8D AF; Type: RECTILINEAR;
#      Focal: 28.0-28.0; Aperture: 2.79999995232-2.79999995232;
#      Crop factor: 1.0; Score: 110)

How to correct lens distortion

import cv2 # OpenCV library

focal_length = 28.0
aperture = 1.4
distance = 10
image_path = '/path/to/image.tiff'
undistorted_image_path = '/path/to/image_undist.tiff'

img = cv2.imread(image_path)
height, width = img.shape[0], img.shape[1]

mod = lensfunpy.Modifier(lens, cam.crop_factor, width, height)
mod.initialize(focal_length, aperture, distance, pixel_format=img.dtype)

undist_coords = mod.apply_geometry_distortion()
img_undistorted = cv2.remap(img, undist_coords, None, cv2.INTER_LANCZOS4)
cv2.imwrite(undistorted_image_path, img_undistorted)

It is also possible to apply the correction via SciPy instead of OpenCV. The lensfunpy.util module contains convenience functions for RGB images which handle both OpenCV and SciPy.

How to correct lens vignetting

Note that the assumption is that the image is in a linear state, i.e., it is not gamma corrected.

import lensfunpy
import imageio

db = lensfun.Database()
cam = db.find_cameras('NIKON CORPORATION', 'NIKON D3S')[0]
lens = db.find_lenses(cam, 'Nikon', 'Nikkor AF 20mm f/2.8D')[0]

# The image is assumed to be in a linearly state.
img = imageio.imread('/path/to/image.tiff')

focal_length = 20
aperture = 4
distance = 10
width = img.shape[1]
height = img.shape[0]

mod = lensfunpy.Modifier(lens, cam.crop_factor, width, height)
mod.initialize(focal_length, aperture, distance, pixel_format=img.dtype)

did_apply = mod.apply_color_modification(img)
if did_apply:
    imageio.imwrite('/path/to/image_corrected.tiff', img)
else:
    print('vignetting not corrected, calibration data missing?')

How to correct lens vignetting and TCA

Note that the assumption is that the image is in a linear state, i.e., it is not gamma corrected. Vignetting should always be corrected first before applying the TCA correction.

import imageio
import cv2
import lensfunpy

db = lensfunpy.Database()
cam = db.find_cameras('Canon', 'Canon EOS 5D Mark IV')[0]
lens = db.find_lenses(cam, 'Sigma', 'Sigma 8mm f/3.5 EX DG circular fisheye')[0]

# The image is assumed to be in a linearly state.
img = imageio.imread('/path/to/image.tiff')

focal_length = 8.0
aperture = 11
distance = 10
width = img.shape[1]
height = img.shape[0]

mod = lensfunpy.Modifier(lens, cam.crop_factor, width, height)
mod.initialize(focal_length, aperture, distance, pixel_format=img.dtype, flags=lensfunpy.ModifyFlags.VIGNETTING | lensfunpy.ModifyFlags.TCA)

# Vignette Correction
mod.apply_color_modification(img)

# TCA Correction
undist_coords = mod.apply_subpixel_distortion()
img[..., 0] = cv2.remap(img[..., 0], undist_coords[..., 0, :], None, cv2.INTER_LANCZOS4)
img[..., 1] = cv2.remap(img[..., 1], undist_coords[..., 1, :], None, cv2.INTER_LANCZOS4)
img[..., 2] = cv2.remap(img[..., 2], undist_coords[..., 2, :], None, cv2.INTER_LANCZOS4)

imageio.imwrite('/path/to/image_corrected.tiff', img)

Installation

Install lensfunpy by running:

pip install lensfunpy

64-bit binary wheels are provided for Linux, macOS, and Windows.

Installation from source on Linux/macOS

If you have the need to use a specific lensfun version or you can't use the provided binary wheels then follow the steps in this section to build lensfunpy from source.

First, install the lensfun library on your system.

On Ubuntu, you can get (an outdated) version with:

sudo apt-get install liblensfun-dev

Or install the latest developer version from the Git repository:

git clone https://github.com/lensfun/lensfun
cd lensfun
cmake .
sudo make install

After that, install lensfunpy using:

git clone https://github.com/letmaik/lensfunpy
cd lensfunpy
pip install numpy cython
pip install .

On Linux, if you get the error "ImportError: liblensfun.so.0: cannot open shared object file: No such file or directory" when trying to use lensfunpy, then do the following:

echo "/usr/local/lib" | sudo tee /etc/ld.so.conf.d/99local.conf
sudo ldconfig

The lensfun library is installed in /usr/local/lib when compiled from source, and apparently this folder is not searched for libraries by default in some Linux distributions. Note that on some systems the installation path may be slightly different, such as /usr/local/lib/x86_64-linux-gnu or /usr/local/lib64.

Installation from source on Windows

These instructions are experimental and support is not provided for them. Typically, there should be no need to build manually since wheels are hosted on PyPI.

You need to have Visual Studio installed to build lensfunpy.

In a PowerShell window:

$env:USE_CONDA = '1'
$env:PYTHON_VERSION = '3.10'
$env:PYTHON_ARCH = 'x86_64'
$env:NUMPY_VERSION = '2.0.*'
git clone https://github.com/letmaik/lensfunpy
cd lensfunpy
.github/scripts/build-windows.ps1

The above will download all build dependencies (including a Python installation) and is fully configured through the four environment variables. Set USE_CONDA = '0' to build within an existing Python environment.