This package implements a PCLines transform for line detection in images.

@INPROCEEDINGS{dubska2011pclines,
    author={M. {Dubská} and A. {Herout} and J. {Havel}},
    booktitle={CVPR 2011},
    title={PClines — Line detection using parallel coordinates},
    year={2011},
}

• Python 3.6+
• numpy
• numba
• scikit-image

The package is on PyPI, so just run following command and install the package.

> pip install pclines

Alternatively, you can download this repository and install manually.

1. Import package

import pclines as pcl

2. Data and observations The observations are 2D weighted coordinates enclosed by a known bounding box. As an example we extract edge points from an image.

image = imread("doc/test.png", as_gray=True)
edges = sobel(image)
r,c = np.nonzero(edges > 0.5)
x = np.array([c,r],"i").T
weights = edges[r,c]

3. Accumulation in PCLines space

h,w = image.shape[:2]
bbox=(0,0,w,h)  #  Bounding box of observations
d = 1024  # Accumulator resolution
P = PCLines(bbox, d) # Create new accumulator
P.insert(x, weights) # Insert observations
p, w = P.find_peaks(min_dist=10, prominence=1.3, t=0.1) # Find local maxima

4. Detected lines

h = P.inverse(p)  # (a,b,c) parameters of lines
X,Y = utils.line_segments_from_homogeneous(h, bbox)  # Convert to line segments for plotting

If you have a suggestion for improvement, let us know by filling an issue. Or you can fork the project and submit a pull request.