it shouldn’t be flipped for one operation and not the other. either it’s flipped, which is then a convolution, or it’s not, so it’s a correlation.
what you mean is that you want its center to be shifted. that’s a matter of taste and not necessarily required inherently by either operation.
some people seem to be doing this shifting business to fix the offset caused by even-sized kernels, where the center doesn’t fall on an element exactly. so for one of erosion/dilation, they go one half left, and for the other operation they go one half right, and that cancels out.
if you want to shift the center, just pass an anchor.
I’d also recommend that you convert your arrays to numpy arrays before calling. makes the code neater.
import numpy as np
import cv2 as cv
el = np.uint8([[0,1,1]])
# this is lopsided, so there's no way around having to think about where the anchor is
im = np.zeros((3,5), dtype=np.uint8)
im[0, 1:2] = 1 # one set
im[1, 1:3] = 1 # two set
im[2, 1:4] = 1 # three set
r = cv.erode(im, el * 255, anchor=(1,0)) # anchor on the center (left 1)
s = cv.dilate(r, el * 255, anchor=(2,0)) # anchor on the right (right 1)
print(im)
print(r)
print(s) # half-element shifts cancel each other out
[[0 1 0 0 0]
[0 1 1 0 0]
[0 1 1 1 0]]
[[0 0 0 0 0]
[0 1 0 0 0]
[0 1 1 0 0]]
[[0 0 0 0 0]
[0 1 1 0 0]
[0 1 1 1 0]]