I want to prove that erosion of a signal by a function $g$ followed by erosion with another function $h$ is equivalent to erosion of the signal by the dilation of the two functions (Erosion property):

$$\left ( f \ominus g \right ) \ominus h = f \ominus \left (g \oplus h \right ) $$

My approach:

By definition of erosion we get that: $$\left ( f \ominus g\right )(x) = ⋀_{y} f(x+y)-g(y)$$

So, we get that $$\left ( f \ominus g \right ) \ominus h = ⋀_{y}\left ( ⋀_{z} f(x+y+z)-g(y)-h(z)\right )$$

Is my approach correct so far? And if yes, how do I go on?


1 Answer 1


Morphological Erosion is not associative as dilation. Much like addition is associative but not subtraction. Consider the following basic definitions first:$$(A \ominus B)^C = A^C \oplus \widetilde{B} \tag{1}$$ $$(A \oplus B)^C = A^C \ominus \widetilde{B} \tag{2}$$

where $\widetilde{B}$ is the reflection of kernel $B$ and $A^C$ is the compliment of image $A$, inverse of image $A$. Now let us evaluate the expression in the question: $$(A \ominus B) \ominus C = ((A\ominus B)^C \oplus \widetilde{C})^C \tag{using eq. 1}$$

$$(A \ominus B) \ominus C = ((A^C\oplus \widetilde{B}) \oplus \widetilde{C})^C \tag{using eq. 1} $$

$$(A \ominus B) \ominus C = (A^C \oplus (\widetilde{B} \oplus \widetilde{C}))^C \tag{since dilation is associative}$$

$$(A \ominus B) \ominus C = A \ominus \widetilde{(\widetilde{B} \oplus \widetilde{C})} \tag{using eq. 2}$$

$$(A \ominus B) \ominus C = A \ominus (B \oplus C) \tag{distributing the reflection}$$

Hope that answers the question

  • $\begingroup$ Thanks for your answer!However, in this case the dilation and erosion operators apply on sets and not on functions, where the definitions are slightly different as far as I am concerned. Could you adjust your answer so that it contains functions $f,g,h$ instead of sets $A,B,C$ , since I don't think I can just simply replace sets with functions. $\endgroup$
    – Rog Fed
    Apr 22, 2020 at 8:46
  • $\begingroup$ Functions are nothing but sets, when you consider their range space or domain as a set of collection of points. For ex: if the range space of a function is A in $\mathbb{R}^N$, the complement set is $\mathbb{R}^N$ -A $\endgroup$ Apr 22, 2020 at 8:55
  • $\begingroup$ In my case,functions represent grayscale images. Can I just replace $A,B,C$ with $f,g,h$ as simple as that: $A \rightarrow f$, $B \rightarrow g$ and $C \rightarrow h$ and be valid with these properties? $\endgroup$
    – Rog Fed
    Apr 22, 2020 at 8:59
  • $\begingroup$ What I have described above is with references to images. That's why I use the terms like reflection of kernel and inverse of Image. It is valid to do that $\endgroup$ Apr 22, 2020 at 9:02
  • $\begingroup$ Yeah, you are right! So, that answers my question. However, do you have any idea how this could be proved by just using the standard definition of erosion between functions, as I wrote down in my question? Because that was the way I approached this problem and I am curious how we can solve it in this case (as an alternative solution). $\endgroup$
    – Rog Fed
    Apr 22, 2020 at 9:08

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.