OpenCV perspective transform in python

Question

I'm trying to rectify an image in python. I have a Homography H (from a rotation matrix that rotates around the x, y and z axis) that looks like this for example: [[ 9.95671447e-01 7.83610423e-02 7.47993630e+02] [ -7.69292630e-02 9.96586377e-01 -4.48354859e+02] [ -3.48494755e-06 1.73615469e-06 9.98300856e-01]]

I thought I could do this woth cv2.perspectiveTransform() but I cant't get it to work. This is the code I use:

   # warp image
   img_cv2 = cv2.imread('surf.jpg', cv2.CV_LOAD_IMAGE_GRAYSCALE)
   # strange output but it does something:
   dst = cv2.perspectiveTransform(img_cv2,H)

But I get the following error:

    Traceback (most recent call last):
    File "C:\directory structure\python_files\Rectification\rectify.py", line 82, in <module>
    dst = cv2.perspectiveTransform(img_cv2,H)
    error: C:\slave\WinInstallerMegaPack\src\opencv\modules\core\src\matmul.cpp:1916: error: (-215) scn + 1 == m.cols && (depth == CV_32F || depth == CV_64F)`</pre>

Can anybody see what is going wrong?

Accept an answer, thus it appears as solved and would help others who come with this same issue. — Anoop K. Prabhu, Feb 25 '14 at 13:33

score 2 · Accepted Answer · answered Feb 20 '14 at 11:26

The source and destination image must be floating point data.

cv2.perspectiveTransform(src, m[, dst]) → dst

Parameters:

src – input two-channel or three-channel floating-point array; each element is a 2D/3D vector to be transformed.

dst – output array of the same size and type as src.

m – 3x3 or 4x4 floating-point transformation matrix.

Refer:http://docs.opencv.org/modules/core/doc/operations_on_arrays.html?highlight=perspectivetransform#cv2.perspectiveTransform

So convert the 8U image to the appropriate datatype.

That's what can be found from your reference: "The function transforms a sparse set of 2D or 3D vectors" - this function is not about transforming images, look at @AldurDisciple answer — Fedor Chervinskii, Oct 11 '16 at 10:43

score 2 · Answer 2 · answered Feb 20 '14 at 11:28

2

I think what you want is cv2.warpPerspective (see documentation (link)), not cv2.perspectiveTransform.

answered Feb 20 '14 at 11:28

BConic

8,750
2
29
55

Abid Rahman K · Answer 3 · 2014-02-20T11:31:59.173

0

It says the image should be np.float32 or np.float64.

So convert the image first by img_cv2 = np.float32(img_cv2).

Then apply the cv2.perspectiveTransform() and cv2.warpPerspective()

Check this tutorial for demo

edited Feb 20 '14 at 11:31

answered Feb 20 '14 at 11:26

Abid Rahman K

51,886
31
146
157

OpenCV perspective transform in python

3 Answers3

Linked