I have an array of bgr values called img_matrix, an empty array called new_img, and another array that tells what index every pixel value in img_matrix should go to in new_img, called img_index. So basically:

for i, point in enumerate(img_index):
    x = point[0]
    y = point[1]
    new_img[y][x] = img_matrix[i]

How can i get rid of the for loop and speed things up? Im sure there's a numpy function that does this.

--some clarification-- my end goal is projecting a 640x480 image from a camera on a drone with a known rotation and displacement, onto the z=0 plane. After projection, the image turns into a grid of points on the z=0 plane resembling a trapezoid. I am trying to "interpolate" these points onto a regular grid. All other methods were too slow (scipy.interpolate, nearest neighbor using k-d tree) so i devised another method. I "round" the coordinates into the closest point on the grid i want to sample, and assign the rgb values of those points to the image matrix new_img where they line up. If nothing lines up, i would like the rgb values to all be zero. If multiple points line up on top of each other, any will do.

an example would maybe be

img_index = 
[[0, 0]
 [0, 1]
 [0, 1]
 [1, 1]]

img_matrix = 
[[1, 2, 3]
 [4, 5, 6]
 [7, 8, 9]
 [10, 11, 12]]

new_img=
[[[1,2,3],[7,8,9]]
 [[0,0,0],[10,11,12]]]

Thanks in advance!

CodePudding user response：

It seems that you want to create an empty array and fill its values at the specific places. You could do it in a vectorised way like so:

img_index = np.array([[0, 0], [0, 1], [0, 1], [1, 1]])
img_matrix = np.array([[1, 2, 3], [4, 5, 6], [7, 8, 9], [10, 11, 12]])
new_img = np.zeros(shape=(2, 2, 3), dtype=int)
new_img[img_index[:,0], img_index[:,1]] = img_matrix
new_img
>>>
array([[[ 1,  2,  3],
        [ 7,  8,  9]],

       [[ 0,  0,  0],
        [10, 11, 12]]])