[scikit-image] Warping image subset
Johannes Schönberger
jsch at demuc.de
Fri Oct 7 02:39:36 EDT 2016
You can do this by iterating over windows of your image with the
following (pseudo-)code:
M = tf.estimate_transform('polynomial',pts2,pts1,order=3)
for minr in range(0, image.shape[0], window_size):
for minc in range(0, image.shape[1], window_size):
coords = M(np.array(np.mgrid[minr:minr+window_size,
minc:minc+window_size]))
transformed_coords = M(coords)
# TODO: reshape transformed_coords into the correct size, refer
to tf.warp_coords for an example.
total_output_image[window_bbox] = warp(image,
transformed_coords, ...)
I hope it is clear what I mean here? I don't have more time at the
moment...
On Fri, Oct 7, 2016, at 02:06 AM, Amaury Dehecq wrote:
> Hello everyone,
>
> I'm using skimage to correct distortions on an image, using a polynomial
> transformation. Basically, my commands can be summarized to:
>
> import skimage.transform as tf
>
> # estimate the transformation matrix from control points before
> (pts2) and after (pts1) distortion
> M=tf.estimate_transform('polynomial',pts2,pts1,order=3)
>
> # warp the initial image im
> warped=tf.warp(im,M)
>
> This does exactly what I want, except that my image is very large (35000
> x 35000) and the script crashes if I try to run it for the whole image.
> So I thought I could just cut my image into subimages and run warp for
> each subimage. But as the matrix indexes are shifted (e.g always in the
> interval [0-5000] instead of [0-350000]), I end up applying the same
> correction to each subimage.
> Is there a way to take into account the fact that my matrix is a subset,
> for example by indicating the actual coordinates instead of using the
> matrix indexes? (Like in the MatLab function imwarp using argument RA if
> that helps understanding)
>
> Thanks a lot,
>
> Amaury
>
> _______________________________________________
> scikit-image mailing list
> scikit-image at python.org
> https://mail.python.org/mailman/listinfo/scikit-image
More information about the scikit-image
mailing list