rebuild image based of numpy arrays

Question

I have this func that splits my initial img into arrays of 100x100

def blockshaped(arr, nrows, ncols):
    h, w = arr.shape
    return (arr.reshape(h//nrows, nrows, -1, ncols)
               .swapaxes(1,2)
               .reshape(-1, nrows, ncols))

Then this loop makes a new_img filled with arrays of zeros if enumerate == even, and block if is not

image = np.full((1000,2000),5)
blocks = blockshaped(image,100,100)

new_image = []
for i,block in enumerate(blocks):
    if i%2==0:
        new_image.append(np.zeros((100,100)))
    else:
        new_image.append(block)

Now I dont know how can I build a final image 1000x2000 filled with my new_image arrays.

Any idea???

tenhjo · Accepted Answer · 2020-09-26 18:59:06Z

1

You can just invert the steps you apply in blockshaped() to achieve this. Note that I renamed some of your variables to make the descriptions shorter.


import numpy as np

def blockshaped(arr, nrows, ncols):
    r, c = arr.shape                                       #    (r, c)
    return (arr.reshape(r//nrows, nrows, c//ncols, ncols)  # -> (r/n, n, c/m, m)
               .swapaxes(1, 2)                             # -> (r/n,  c/m, n, m)
               .reshape(-1, nrows, ncols))                 # -> (r/n * c/m, n, m)

r, c = 1000, 2000
n, m = 100, 100
image = np.full((r, c), 5)
blocks = blockshaped(image, nrows=n, ncols=m)

# Create new empty image and only fill in the even blocks
new_image = np.zeros((len(blocks), n, m))
new_image[::2] = blocks[::2]
# new_image[1::2] = np.full((n, m), 13)  # do something odd if you want

# reverse blockshaped                              #    (r/n * c/m, n, m)
new_image = (new_image.reshape(r//n, c//m, n,  m)  # -> (r/n,  c/m, n, m)
                      .swapaxes(1, 2)              # -> (r/n, n, c/m, m)
                      .reshape(image.shape))       # -> (r, c)

import matplotlib.pyplot as plt
plt.imshow(new_image)

edited Sep 26, 2020 at 18:59

answered Sep 26, 2020 at 18:46

tenhjo

4,5911 gold badge17 silver badges48 bronze badges

Sign up to request clarification or add additional context in comments.

1 Comment

Agustin Gonzalez Ribas Over a year ago

it works fine, but it was my fault I placed the example of ever or odd just to not complicate too much the code, in fact depending of a condition, the block can be zeros or not, So I end with an arr that contain some original blocks and some zero blocks. And I am confused when I want to join all this blocks.

Ahmet · Accepted Answer · 2020-09-26 19:23:38Z

You have created 200 number of 100 x 100 array using

image = np.full((1000, 2000), 5)
blocks = block_shaped(image, 100, 100)

Then you want to fill the new array with the block values if counter is odd number.

new_image = []
for i,block in enumerate(blocks):
    if i%2==0:
        new_image.append(np.zeros((100,100)))
    else:
        new_image.append(block)

Instead of initializing new_image as an empty list, initialize as a numpy zeros array. The problem is what is the shape of the new_image?

The answer depends.

If you want each row to be the blocks value, then you need to declare the size:

new_image = np.zeros(shape=(2000, 10000))

Since each block is 100 x 100 = 10000 and the size of block is 200 x 100 x 100. We place each row block value, for instance new_image[0, 10000] = block.

import numpy as np
from PIL import Image


def block_shaped(arr, n_rows, n_cols):
     h, w = arr.shape
     return (arr.reshape(h//n_rows, n_rows, -1, n_cols).swapaxes(1, 2).reshape(-1, n_rows, n_cols))


image = np.full((1000, 2000), 5)
blocks = block_shaped(image, 100, 100)

new_image = np.zeros(shape=(2000, 10000), dtype=np.uint8)

for i, block in enumerate(blocks):
     if i % 2 != 0:
         new_image[i, :] = block.flatten()

new_image = Image.fromarray(new_image)
new_image.show()

Result will be a blank image, but if you change image = np.full((1000, 2000), 5) to image = np.full((1000, 2000), 155). Then the result will be:
- Upper part becomes gray.

If you want to set only (100, 100) parts of the image to the block for each odd i:

import numpy as np
from PIL import Image


def block_shaped(arr, n_rows, n_cols):
    h, w = arr.shape
    return (arr.reshape(h//n_rows, n_rows, -1, n_cols).swapaxes(1, 2).reshape(-1, n_rows, n_cols))


image = np.full((1000, 2000), 155)
blocks = block_shaped(image, 100, 100)

new_image = np.zeros(shape=(1000, 2000), dtype=np.uint8)

for i, block in enumerate(blocks):
    if i % 2 != 0:
        new_image[i:100+i, i:100+i] = block

new_image = Image.fromarray(new_image)
new_image.show()

Result:

Thanks for the answer, I think I have explain the problen wrong. 1.- I have an image (2000x1000) 2.- I split this img in sections of (100x100) total =200 blocks 3.- I loop all blocks and depending of a condition the block is left untouched, or is transformed into np.zeros 4,.- I end with a arr with some blocks untoched and some blocks with zeros. 5.- All this 4 steps are achived, but now I am confused about how can I rejoin all these blocks in a 2000x1000 img Thanks for your time!

Collectives™ on Stack Overflow

rebuild image based of numpy arrays

2 Answers 2

1 Comment

1 Comment

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

2 Answers 2

1 Comment

1 Comment

Your Answer

Sign up or log in

Post as a guest

Related