Perspective transform question

[ktenney]

I am having problems with Image.transform method=Image.Transform.PERSPECTIVE
documentation seems kind of scarce.

I thought the 'data' arg was a sequence of 8 (x,y) values indicating the corners of source and destination image, however this is not working.

Any suggestions appreciated.

Thanks,
Kent

[radarhere]

The official documentation is at https://pillow.readthedocs.io/en/stable/reference/Image.html#PIL.Image.Image.transform

When you say it's not working, what do you mean exactly? Here is code that will rotate one of our test images by 90 degrees.

import math
from PIL import Image
im = Image.open("hopper.png")

angle = -math.radians(90)
matrix = [
    round(math.cos(angle), 15),
    round(math.sin(angle), 15),
    0.0,
    round(-math.sin(angle), 15),
    round(math.cos(angle), 15),
    0.0,
    0,
    0,
]
matrix[2] = (1 - matrix[0] - matrix[1]) * im.width / 2
matrix[5] = (1 - matrix[3] - matrix[4]) * im.height / 2

im.transform(im.size, Image.Transform.PERSPECTIVE, matrix).save("rotated_90.png")

[ktenney]

ok, I was thinking of something available in cv2:
https://theailearner.com/tag/cv2-getperspectivetransform/
which takes outline points of src and dst images, returns
the correct matrix for cv2.warpPerspective

It doesn't look like Pillow offers this aid to generating a matrix.

Is the matrix returned by cv2.getPerspectiveTransform
the same as expected by the Pillow Image.transform?

Thanks,
Kent

[radarhere]

In my earlier code, matrix was [0.0, -1.0, 128.0, 1.0, 0.0, 0.0, 0, 0]

If I ask cv2.getPerspectiveTransform to rotate the image,

import cv2
import numpy
print(cv2.getPerspectiveTransform(
  numpy.float32([(0, 0), (128, 0), (128, 128), (0, 128)]),
  numpy.float32([(128, 0), (128, 128), (0, 128), (0, 0)])
))

I get

[[  0.  -1. 128.]
 [  1.   0.   0.]
 [ -0.   0.   1.]]

The only difference is that the cv2 matrix is not 8 points, but 3x3 with a 1 in the 9th slot.

[ktenney]

great, this is extremely helpful!

Thanks,
Kent