Laplacian Filter kernel size meaning

Question

I'm using laplacian filter in opencv.

I usually use laplacian filter with kernel size 3 which is default setting in opencv.

I'm just curious what happen(intuitively) when I choose bigger size kernel?

I think if I take small size such as 3x3, this gives me 3x3 matrix. So the result is "how much change in pixel intensity in 3x3 neighbors.

However if I take bigger size such as 7x7, this gives me 7x7 matrix. So the result is "how much change in pixel intensity in 7x7 neighbors.

So, intuitively I think small kernel size gives me small local change, on the other hand big kernel size gives me broader change.

Is is right?

Answer 1

Internally, the Laplacian() operator applies a Sobel filter and the kernel_size parameter specifies the dimensionality of that filter. Therefore, your intuition about applying the filter to a neighborhood proportional to the kernel_size is correct. Consider plotting the results of applying Laplacian() with various kernel_size values side by side if you want to solidify your intuition. I also recommend you look at OpenCV's documentation as it contains all the theory you need. I hope that helps!