How to extract a sub-array in numpy efficiently?

Question

I have an image given as a 3D numpy array (width, height and color as dimensions). Now I would like to extract all possible sub-images of a given shape. For example, the width and height of the original image is 300 and 200 pixels, respectively, and I would like to extract all sub-images with width and heights equal to 10 and 20 pixels, respectively. Moreover, each sub-image I would like to have as 1D array (for that I need to specify a particular ordering of pixels).

I solve the problem in the following way:

for col0 in range(w_max - max_shift):
    x3s_new = [x + col0 for x in x3s]
    for row0 in range(h_max - max_shift):

        vec_1 = []
        for col_shift, row_shift in px_inds:
            col = col0 + col_shift
            row = row0 + row_shift
            vec_1 += [ia[row, col, 0], ia[row, col, 1], ia[row, col, 2]]

        y3s_new = [y + row0 for y in y3s]
        vec_2 = list(ia[y3s_new, x3s_new, z3s])

In the above code I make a loop over columns and rows of the "matrix" representing the image. Then each pixel (given by its column and row) I treat as the left-top corner of the sub-image and extract the sub-image.

The vec_1 and vec_2 are the desired sub-images given as 1D arrays (lists). They are identical, I just wanted to test what way is faster. Surprisingly, it take more time to generate vec_2 than vec_1. However, according to this answer to my question it was expected to be faster. So, why it is not faster?

Finally, I also would like to know if there is a faster alternative to looping over all columns and rows of the image-matrix.

To summarize, my question is: How to achive what I need in faster way? At the moment it takes me about 5 minutes to "process" one image and it is not acceptable for my purposes.

Do you know about slicing?

cfh
– cfh

2015-05-19 06:18:20 +00:00
Commented May 19, 2015 at 6:18 — cfh
– cfh, Commented May 19, 2015 at 6:18
@cfh, no (some_symbols_to_make_it_acceptably_long).

Roman
– Roman

2015-05-19 07:19:38 +00:00
Commented May 19, 2015 at 7:19 — Roman
– Roman, Commented May 19, 2015 at 7:19

rth · Accepted Answer · 2015-05-19 09:03:29Z

5

You should use Numpy slicing.

Given a 3D numpy array M, you can select an sub-array of it, with for instance,

M_selection = M[i_min:i_max, j_min:j_max, k_min:k_max]

or alternatively, explicitly defining the slice,

sl_i = slice(i_min, i_max)
sl_j = slice(j_min, j_max)
sl_k = slice(k_min, k_max)
M_selection = M[sl_i, sl_j, sl_k]

where (i_min, i_max), etc. are the boundaries of the sub-array.

See the documentation of advanced Numpy indexing for more details.

answered May 19, 2015 at 9:03

rth

11.3k7 gold badges58 silver badges88 bronze badges

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

Collectives™ on Stack Overflow

How to extract a sub-array in numpy efficiently?

1 Answer 1

Comments

Your Answer

Linked

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

Comments

Your Answer

Sign up or log in

Post as a guest

Linked

Related