Python Sobel implementation ignores defined kernel size #893
Description
Hello,
When I was using GRIP's Sobel edge detection, I found that the Mat returned from the Sobel I saw on GRIP did not match what the generated Python code created. Instead, the Mat would always have a ksize = 3 (the default value) no matter what kernel size I set.
This is what appeared in GRIP:
This is what appeared in Python:
The bottom image (the Sobel edge) exactly matches the Mat I see in GRIP when I use kernel size 3.
Laplacian doesn't have this issue however.
The solution I found:
Change this -
def __cv_sobel(src, dx, dy, k_size, scale, delta, border_type):
"""Find edges by calculating the requested derivative order for the given image.
Args:
src: A numpy.ndarray.
k_size: Odd number that is size of the kernel.
scale: Scaling factor for Sobel.
delta: Offset for values in Sobel.
border_type: Opencv enum.
Returns:
The result as a numpy.ndarray.
"""
return cv2.Sobel(src, 0, (int)(dx + 0.5), (int)(dy + 0.5), (int)(k_size + 0.5),
scale = scale, delta = delta, borderType = border_type)
to this.
def __cv_sobel(src, dx, dy, k_size, scale, delta, border_type):
"""Find edges by calculating the requested derivative order for the given image.
Args:
src: A numpy.ndarray.
k_size: Odd number that is size of the kernel.
scale: Scaling factor for Sobel.
delta: Offset for values in Sobel.
border_type: Opencv enum.
Returns:
The result as a numpy.ndarray.
"""
return cv2.Sobel(src, 0, (int)(dx + 0.5), (int)(dy + 0.5), ksize=(int)(k_size + 0.5),
scale = scale, delta = delta, borderType = border_type)
Adding a "ksize=" to directly specify the ksize works (this is what the Laplacian function does). Directly specifying the ksize sets the kernel size correctly.