Skip to content

Instantly share code, notes, and snippets.

What would you like to do?
# python input.jpeg output.tif
# n-band image -> PCA -> n-band TIFF image
# with lots of hackety assumptions
# (e.g., output is same type as input)
from sys import argv
import rasterio as rio
import numpy as np
from sklearn import decomposition
with[1]) as src:
meta = src.meta
pixels =
dtype = meta['dtype']
count = meta['count']
# Todo: make this cleaner:
pixels = np.dstack([c.ravel() for c in pixels])[0]
pca = decomposition.PCA(n_components=count, whiten=False)
for band in range(len(pca.components_)):
'Band {0} will hold {1:.4g} of variance with weights:\n{2}'.format(
', '.join("{0:.4g}".format(x) for x in pca.components_[band])))
# .format() within .format()! Wow! Very pro move, very well
# respected technique, 9.7 even from the East German judges!
# Here's the actual work:
out = pca.transform(pixels)
# This is the messy reverse of the messy ravel above:
xyz = np.array([
out[:,c].reshape(meta['height'], meta['width'])
for c in range(count)
# Scale each band separately to fill out the data type.
# (You either really want this or really don't want this.)
xyz = np.array([
((c - np.amin(c))/(np.amax(c) - np.amin(c)))*np.iinfo(dtype).max
for c in xyz
xyz = xyz.astype(dtype)
'transform': meta['affine'],
'driver': 'GTiff'
with[2], 'w', **meta) as dst:
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
You can’t perform that action at this time.