System Information

Python 3.10.16
Ubuntu 24.04 LTS
opencv-python 4.11.0.86

Detailed description

getPerspectiveTransform produces wrong output on certain data. In the code below, these different ways to get inv_M should all be able to correctly transfom dst_point to src_point, but the result (src_new) from inv_M1 is incorrect. This only happens when diff is 320. This issue is probably related to #11944 (in yr 2018). But this bug is still in the latest opencv-python.

Steps to reproduce

import cv2
import numpy as np

src_points = np.float32([[0, 0], [1280, 0], [0, 1024], [1280, 1024]])

diff = 320
dst_points = np.float32([[diff, 512], [1280-diff, 512], [0, 1024], [1280, 1024]])

M = cv2.getPerspectiveTransform(src_points, dst_points)
inv_M1 = cv2.getPerspectiveTransform(dst_points, src_points)
inv_M2 = np.linalg.inv(M)
inv_M3 = cv2.findHomography(dst_points, src_points)[0]

np.set_printoptions(suppress=True)

# print(inv_M1)
# print(inv_M2)
# print(inv_M3)

for inv_M in [inv_M1, inv_M2, inv_M3]:
    dst_new = np.concatenate([dst_points, np.ones((4, 1))], axis=1)
    src_new = np.matmul(inv_M, dst_new.T).T
    src_new = src_new[:, :2] / src_new[:, 2].reshape(-1, 1)

    # the output should all be the same to src_points
    print("src_new:")
    print(src_new)

Issue submission checklist

• I report the issue, it's not a question
• I checked the problem with documentation, FAQ, open issues, forum.opencv.org, Stack Overflow, etc and have not found any solution
• I updated to the latest OpenCV version and the issue is still there
• There is reproducer code and related data files (videos, images, onnx, etc)