ptcfextract computes per-point local geometry features from point clouds stored in LAS/LAZ, useful for feature extraction research pipelines:

• Linearity
• Planarity
• Scattering

It can be used as:

• a library (feature extraction only, integrate with your own pipeline)
• a ready-to-run pipeline (downsample/outlier removal + visualization + two outputs)
• a CLI tool (ptcfextract)

For each point, neighbors within a radius are used to compute a covariance matrix and eigenvalues. The outputs are:

• linearity = (e2 - e1) / e2
• planarity = (e1 - e0) / e2
• scattering = e0 / e2

Robust handling:

• If a point has too few neighbors → outputs are NaN
• If covariance is degenerate or division by zero would occur → outputs are NaN
• Small negative eigenvalues due to numeric noise are clipped to 0

If your output base is out.laz, the tool writes:

out_rgbLPS_overwriteRGB.laz

• RGB is overwritten with normalized [L,P,S]
  • R = Linearity
  • G = Planarity
  • B = Scattering
• All other original point dimensions are preserved

This is for instant visualization in CloudCompare / other viewers.

out_addLPS_extraFields_keepOriginalRGB.laz

• Original RGB remains unchanged
• Adds three ExtraBytes scalar fields (float32):
  • linearity
  • planarity
  • scattering
• All other original point dimensions are preserved

This is for analysis workflows where you want to keep original appearance.

Note: If the input has no RGB fields and you request Output 1, ptcfextract upgrades point format to write RGB and prints a warning (some vendor-specific extra dims may not be preserved).

pip install ptcfextract

pip install "ptcfextract[laz]"

ptcfextract -i input.laz -o out.laz

ptcfextract --help

ptcfextract -i input.laz -o out.laz --radius 0.5 --min-neighbors 15

ptcfextract -i input.laz -o out.laz --downsample --voxel 0.1

ptcfextract -i input.laz -o out.laz --outlier none

ptcfextract -i input.laz -o out.laz --outlier sor --sor-nn 20 --sor-std 2.0

ptcfextract -i input.laz -o out.laz --outlier ror --ror-np 6 --ror-r 0.5

ptcfextract -i input.laz -o out.laz --no-viz

ptcfextract -i input.laz -o out.laz --viz linearity
ptcfextract -i input.laz -o out.laz --viz planarity
ptcfextract -i input.laz -o out.laz --viz scattering

ptcfextract -i input.laz -o out.laz --viz rgb

ptcfextract -i input.laz -o out.laz --no-overwrite-rgb

ptcfextract -i input.laz -o out.laz --no-add-dims

Use this if you want your own:

• downsampling strategy (e.g., random, uniform, octree)
• visualization (CloudCompare, PyVista, Open3D custom)
• ML pipeline / feature engineering

import laspy
import numpy as np
from ptcfextract import compute_lps_features

las = laspy.read("input.laz")
xyz = np.vstack([las.x, las.y, las.z]).T

features = compute_lps_features(xyz, radius=0.5, min_neighbors=10)

linearity = features[:, 0]
planarity = features[:, 1]
scattering = features[:, 2]

import laspy
import numpy as np
from ptcfextract import voxel_downsample_indices, compute_lps_features

las = laspy.read("input.laz")
xyz = np.vstack([las.x, las.y, las.z]).T

idx = voxel_downsample_indices(xyz, voxel_size=0.1)
xyz_ds = xyz[idx]

features_ds = compute_lps_features(xyz_ds, radius=0.5, min_neighbors=10)

import laspy
import numpy as np
from ptcfextract import outlier_removal_indices, compute_lps_features

las = laspy.read("input.laz")
xyz = np.vstack([las.x, las.y, las.z]).T

inliers = outlier_removal_indices(
    xyz,
    mode="sor",
    sor_nb_neighbors=20,
    sor_std_ratio=2.0,
    ror_nb_points=6,
    ror_radius=0.5,
)

xyz_in = xyz[inliers]
features_in = compute_lps_features(xyz_in, radius=0.5, min_neighbors=10)

from ptcfextract import features_to_rgb

rgb01 = features_to_rgb(features_in)   # Nx3 float in [0,1]

from ptcfextract import scalar_to_colormap_rgb

rgb_L = scalar_to_colormap_rgb(features_in[:, 0])  # linearity colormap
rgb_P = scalar_to_colormap_rgb(features_in[:, 1])  # planarity colormap
rgb_S = scalar_to_colormap_rgb(features_in[:, 2])  # scattering colormap

from ptcfextract import process_pointcloud

process_pointcloud(
    input_path="input.laz",
    output_base="out.laz",
    radius=0.5,
    min_neighbors=10,
    use_downsample=True,
    voxel_size=0.1,
    outlier_mode="sor",
    sor_nb_neighbors=20,
    sor_std_ratio=2.0,
    visualize=True,
    visualize_mode="rgb",
    export_rgb_lps_overwrite=True,
    export_add_dims_keep_original_rgb=True,
)

If you computed features yourself and want to export using the same preservation logic:

import laspy
import numpy as np
from ptcfextract import write_las_preserve_all, compute_lps_features, features_to_rgb

base = laspy.read("input.laz")
xyz = np.vstack([base.x, base.y, base.z]).T
features = compute_lps_features(xyz, radius=0.5, min_neighbors=10)
rgb01 = features_to_rgb(features)

orig_idx = np.arange(xyz.shape[0], dtype=np.int64)

# Output 1: overwrite RGB with LPS
write_las_preserve_all(
    out_path="out_rgbLPS_overwriteRGB.laz",
    base_las=base,
    kept_orig_indices=orig_idx,
    rgb01=rgb01,
    add_lps_dims=False,
)

# Output 2: keep original RGB, add extra dims
write_las_preserve_all(
    out_path="out_addLPS_extraFields_keepOriginalRGB.laz",
    base_las=base,
    kept_orig_indices=orig_idx,
    rgb01=None,
    add_lps_dims=True,
    lps_features=features,
)

• radius, voxel, and ror-r are in the same units as the LAS coordinates.
• If your cloud is in feet, typical values often need to be larger than meter-based defaults.

You’ll see a NaN count printed (pipeline mode). If it’s high:

• Increase --radius
• Decrease --min-neighbors
• Downsample less aggressively (smaller --voxel)
• If the cloud is sparse, a small radius may not find enough neighbors

• Enable downsampling (--downsample --voxel 0.1 or larger)
• Disable visualization (--no-viz)
• Consider outlier removal only after downsampling

• Output 2 still works (extra dims added, original dims preserved).

• Output 1 requires RGB, so ptcfextract upgrades point format to write RGB and prints a warning.

  • This can drop some unusual vendor-specific dimensions depending on the LAS point format and library limitations.

• Output 1: visualize using RGB.

• Output 2: visualize using scalar fields:

  • linearity, planarity, scattering
  • In CloudCompare: open the scalar field dropdown to choose which to view.

If you use ptcfextract in academic work, cite the Zenodo DOI associated with the GitHub release.

"# ptcfextract"