如何使用透视变换分割 openCV 中的白色钢琴键

Question

I have a perspective image of a piano keyboard from the left side, like this:

I manually outline the four corners of the keyboard and fed it into the openCV getPerspectiveTransform function to get a overhead view of the keyboard. The result is this:

overhead view

I've found that if I do not outline the 4 corners of the keyboard exactly, the edges of the white keys in the perspective transform output will not be exactly vertical.

Then I want to produce an outline of all the white keys. Since the white keys of the piano are all evenly spaced, I just divided the top view image equally into 33 parts horizontally. Then I scaled the bottom edges of the outline by the ratio of the length of the bottom side to the top side of the keyboard in the original image. I got this output.

However, the outlines of the white keys do not match the white key edges.

How do I take advantage of the fact that the white keys are all evenly spaced to segment the white piano keys, even when the edges of the white keys are not perfectly vertical?

*I don't want to use edge detection algorithm to segment the white keys as sometimes in certain lighting conditions, the edges are not visible.

Answer 1

It looks like the perspective transformation is a bit off using only 4 points. Try using a few more and use cv2.findHomography together with cv2.warpPerspective to do the job. Once you have a better perspective transformation, you can continue doing the rest.

Code sample of using v2.findHomography together with cv2.warpPerspective can be found:

https://www.learnopencv.com/homography-examples-using-opencv-python-c/

Good luck.