Pergunta

In our lectures at the university, we got following definition for Correlation with a Kernel K with dimension n:

sum of sum of K(i,j) * I(x+i, y+j), where i,j goes from -n to n.

Convolution is defined as follows:

sum of sum of K(i,j) * I(x-i, y-j), where i,j goes from -n to n.

However, looking at the animation here:

The way they do the convolution is how correlation is defined.

Whats going on here? Is the definition given at the lectures wrong?

Foi útil?

Solução 2

I consulted the professor. Turns out the animation in Wikipedia is wrong. Even though thanks to the symmetry the result is the same.

Outras dicas

The definition is absolutely correct.

Important point to note here is, both correlation and convolution are identical only when the filter I is symmetric.

Licenciado em: CC-BY-SA com atribuição
Não afiliado a StackOverflow
scroll top