Elastic transformation of an image in Python

elastic_transform.py

import numpy as np
from scipy.ndimage.interpolation import map_coordinates
from scipy.ndimage.filters import gaussian_filter

def elastic_transform(image, alpha, sigma, random_state=None):
    """Elastic deformation of images as described in [Simard2003]_.


    .. [Simard2003] Simard, Steinkraus and Platt, "Best Practices for
       Convolutional Neural Networks applied to Visual Document Analysis", in
       Proc. of the International Conference on Document Analysis and
       Recognition, 2003.
    """
    assert len(image.shape)==2

    if random_state is None:
        random_state = np.random.RandomState(None)

    shape = image.shape

    dx = gaussian_filter((random_state.rand(*shape) * 2 - 1), sigma, mode="constant", cval=0) * alpha
    dy = gaussian_filter((random_state.rand(*shape) * 2 - 1), sigma, mode="constant", cval=0) * alpha

    x, y = np.meshgrid(np.arange(shape[0]), np.arange(shape[1]), indexing='ij')
    indices = np.reshape(x+dx, (-1, 1)), np.reshape(y+dy, (-1, 1))
    
    return map_coordinates(image, indices, order=1).reshape(shape)

hey, just wondering what code did you use to load in the image?

You can load using

from skiamge import io
img = io.imread(filename)

Thanks for sharing. However an image is usually a 3-d array. This code seems to require the image is a 2-d array?

You can add a loop to achieve 3-D image.

Sorry about my poor knowledge, but you are using gaussian filter for what exactly?

Sorry about my poor knowledge, but you are using gaussian filter for what exactly?

A c++ implementation can be found here Implementation of elastic distortion algorithm in C++

I wrote the reason why we use gaussian filter.

Sorry about my question, but how to show the transformed image ?

for those who have trouble with using this code
first, the image has to be square like: (x,x,3)
second, for showing the transformed image ex: transformed_image=elastic_transform(img,40,2) then using plt.imshow(transformed_image)

Sorry about my poor knowledge, but you are using gaussian filter for what exactly?

It is used to create the random deformations dx and dy. If you didn't have dx and dy, then the deformed mesh grid would be the same as the un-deformed mesh grid - so no deformation would happen

for those who have trouble with using this code
first, the image has to be square like: (x,x,3)
second, for showing the transformed image ex: transformed_image=elastic_transform(img,40,2) then using plt.imshow(transformed_image)

1. The image does not have to be square
2. The image has to be grayscale, i.e. the shape has to be (a,b) - this line of code makes sure of that assert len(image.shape)==2