CODEWITHSUNDEEP
pythonnumpy
Steven
Steven

Reputation: 1173

Numpy rot90 for 2d points

I have an image and several 2D points [(x, y), ...]. I would like to rotate both the image and the points by 90 degrees.

For the image I can simply use np.rot90. For the points I figured I have to rotate them 90 degrees around the center of the image to make them align correctly (for this I used https://stackoverflow.com/a/58781388/2445065)

Original image (note: red point is not part of the image):

enter image description here

After rotating the image and the point:

enter image description here

Whereas I would expect the red dot to be at the bottom left corner of the yellow square... Does anyone know what I'm doing wrong? Thanks in advance

Upvotes: 0

Views: 883

Answers (2)

Mendi Barel
Mendi Barel

Reputation: 3677

This python function will do rot90 for points:

def rot90points(x, y, k, hw):
    """
        a = np.zeros((5, 12), dtype="uint8");
        x, y = np.array([2, 5, 7]), np.array([4, 4, 2])
        a[y, x] = 200;
        for k in range(10):  # test
            a1 = np.rot90(a, k);
            x1, y1 = rot90points(x, y, k, a.shape);
            assert np.all(a1[y1, x1] == 200)
    """
    k = k % 4
    if k == 0: return x, y
    elif k == 1: return y, hw[1] - 1 - x
    elif k == 2: return hw[1] - 1 - x, hw[0] - 1 - y
    elif k == 3: return hw[0] - 1 - y, x
    else: raise ValueError(f"k error {k}")

Upvotes: 1

Steven
Steven

Reputation: 1173

I finally found it:

To align the points and additional translation is require. np.rot90 does not equal a translation around the center but rather a rotation around the center PLUS a translation such that the top left corner of the rotated image is located at (0, 0).

You can get the translation by subtracting |origin_rotated.x - origin_original.x| from the x-coordinate and |origin_rotated.y - origin_original.y| from the y-coordinate.

Upvotes: 1

Related Questions