CODEWITHSUNDEEP
pythonopencvimage-processingpython-imaging-library
The Oracle
The Oracle

Reputation: 488

How to vectorize an image using python

Is there a way to convert an image into a vectorized form such as as follows:

I have looked this up searching for CNN, Pillow and CV2 methods however I didn't find any available resource online to implement this image transformation.

Appreciating your help on this matter.

Normal Image:

enter image description here

Vectorized Image:

enter image description here

Update:

I have implemented image segmentation using SLIC method, the following is new code and Resultant Image. As can be seen, the new image is close to the vectorized image(desired outcome) however the quality/detail difference is still alot.

Any further ideas?

from skimage import data, segmentation, color
from skimage.future import graph
from matplotlib import pyplot as plt
from PIL import Image
from skimage import io

path='/home/user/Desktop/Image1.jpeg'
img = io.imread(path)

labels1 = segmentation.slic(img, compactness=50, n_segments=5000,
                        start_label=1)
out1 = color.label2rgb(labels1, img, kind='avg', bg_label=0)
out1 = Image.fromarray(out1, 'RGB')
out1.save('/home/user/Desktop/1.png')
g = graph.rag_mean_color(img, labels1, mode='similarity')
labels2 = graph.cut_normalized(labels1, g)
out2 = color.label2rgb(labels2, img, kind='avg', bg_label=0)
out2 = Image.fromarray(out2, 'RGB')
out2.save('/home/user/Desktop/2.png')

fig, ax = plt.subplots(nrows=2, sharex=True, sharey=True, figsize=(6, 8))

ax[0].imshow(out1)
ax[1].imshow(out2)

for a in ax:
    a.axis('off')

plt.tight_layout()
plt.show()

New Image(N_Segments=2000)

enter image description here

New Image (N_segments=5000, took 5 minutes of computational time) enter image description here

Upvotes: 7

Views: 13099

Answers (1)

Ebrahim Badrestani
Ebrahim Badrestani

Reputation: 86

For those who want a simple and fast python solution:

from matplotlib.image import imread, imsave
from scipy import ndimage
from sklearn.cluster import KMeans


def segment_image(image, n_segments, smoothing_sigma):
    smoothed = ndimage.gaussian_filter(image, (smoothing_sigma, smoothing_sigma, 0))
    kmeans = KMeans(n_segments)
    segment_ids = kmeans.fit_predict(smoothed.reshape(-1, 3)).reshape(*image.shape[:2])
    new_image = kmeans.cluster_centers_.astype('uint8')[segment_ids]
    return new_image


image = imread('yCMXP.jpg')
segmented = segment_image(image, 8, 3)
imsave('segmented.png', segmented)

The result:

enter image description here

The quality is a bit lower than the expected result but it gets the job done by a fast and extendable python script.

Upvotes: 0

Related Questions