`Laplacian` is incorrect

[cospectrum]

The actual Laplacian can have negative values, and for a GrayImage the max value can be much higher than 255. But the merged version https://github.com/image-rs/imageproc/pull/516 doesn't care about it. So I want to remove the detector and at least fix the filter. Also approximation of Laplacian is not documented, and OpenCV uses [0, 1, 0, 1, -4, 1, 0, 1, 0] by default.

[thecodergus]

I read your points, reviewed the literature I consulted a year ago (Digital Image Processing, Global Edition - Rafael C. Gonzalez and Richard E. Woods) and I say that it is correct as it is. The default value of OpenCV is the adjacency transform, if you read my previous constructor, the use of [0, 1, 0, 1, -4, 1, 0, 1, 0] was present as an optional situation (The result is "dirtier"), if OpenCV uses it as default, it was a decision of the developers, in the literature this is a secondary formula. As for what you mentioned about the real Laplacian being able to have negative values, I didn't understand your point very well, in a GrayImage the resulting value will never be greater than 255 or negative.

[cospectrum]

In GrayImage they will not be negative or larger than 255, since in this format the pixels are stored as u8. But it doesn't mean that values are correct.

[cospectrum]

If the input values are in [0, 255], then after Laplacian (with [0, 1, 0, 1, -4, 1, 0, 1, 0] kernel) they can be in [-1020, 1020].

[cospectrum]

See horizontal_sobel, for example. The input is GrayImage, but the output is Image<Luma<i16>> to make sure we don't lose the value.

[thecodergus]

See horizontal_sobel, for example. The input is GrayImage, but the output is Image<Luma<i16>> to make sure we don't lose the value.

I understand your point, but for what the algorithm proposes and does is correct, no information will be lost

[cospectrum]

Do you think laplacian_filter is correct? It's definitely not. Because I have variance of laplacian and laplacian of gaussian in the production, and current implementation from imageproc master just not gonna work for me, so I will continue to use my own implementation, which does not lose value.

[cospectrum]

https://docs.opencv.org/4.9.0/d5/db5/tutorial_laplace_operator.html

[thecodergus]

https://docs.opencv.org/4.9.0/d5/db5/tutorial_laplace_operator.html

I understand, I removed my suggestion about this, the only point that really bothered me was the removal of the function