4
$\begingroup$

I am attempting to implement the Gabor Descriptor based on the following paper Image Copy Detection Using a Robust Gabor Texture Descriptor. It is the best paper that I have found on the subject. It is not perfect, as it has some ordering issues with some equations and paragraphs that only make sense after later sections. No biggie, it is only a matter of time consumed to decipher the intent.

What is stopping me right now, is that one of the parameters is never defined in the paper. It first appears in $(1)$ of Section 2. There we have the following factor: $e^{2\pi jWx}$

$W$ is later assumed to be 1 and $x$ is the parameter of the function, but $j$ is never defined (though it is used again in several subsequent equations, especially in section 3.1, of most interest is $(11)$).

So I am seeking either a definition of $j$. A nice bonus would be a better paper (or book) defining the Gabor Descriptor.

Improve this question
$\endgroup$

2 Answers 2

Reset to default
5
$\begingroup$

In this context, $j$ is the imaginary unit: $j = \sqrt{-1}$. The complex exponential function represents what they call an "oriented sinusoidal grating." The fact that $j$ is the imaginary unit in this case can be verified by noticing the relationship between $(1)$ and $(2)$ in Section 2; this is an example of the frequency-shifting property of the Fourier transform. That is, multiplying the Gaussian function $h(x,y)$ by the complex exponential function results in a shift in the Fourier transform of the result along the axis that the sinusoid runs across (the $x$ axis in your example).

Improve this answer
$\endgroup$
4
  • $\begingroup$ I have never seen $j$ used for the imaginary unit, always been $i$. Didn't even occur to me. Though I'm now somewhat confused. What is the point of that expansion? Doesn't Euler formula tell us that $e^{2 \pi i} = 1$? So $e^{2\pi i W x} = (e^{2\pi i})^{Wx}=1$? $\endgroup$
    – Alex K
    Commented Feb 18, 2012 at 21:16
  • $\begingroup$ $j$ is used as the imaginary unit in many engineering contexts, especially in electrical engineering (as Hilmar pointed out below, since $i$ is used as a variable for electrical current). In your example, if $Wx$ is not an integer, then $(e^{j 2\pi})^{Wx}$ is a root of $1$, which is not necessarily equal to $1$. For example, if $Wx = 0.5$, then $(e^{j 2\pi})^{Wx} = -1$. $\endgroup$
    – Jason R
    Commented Feb 18, 2012 at 21:27
  • $\begingroup$ I don't think I follow. $e^{2\pi i} = 1$ by Euler. So, it's $(1)^{Wx}$. 1 to any real power is 1, is it not? $\endgroup$
    – Alex K
    Commented Feb 18, 2012 at 21:39
  • 1
    $\begingroup$ "So, it's $(1)^{Wx}$. $1$ to any real power is $1$, is it not?" Have you ever considered that $1^{1/2}$ might equal $-1$? $\endgroup$
    – Dilip Sarwate
    Commented Feb 20, 2012 at 4:32
5
$\begingroup$

Electrical Engineers typically use $j$ for the imaginary unit since $i$ is reserved for current.

The whole terms is a complex exponential, i.e. it's magnitude is always one and the period is W*x. It's only unity if W*x is an integer.

The identity you use only works if W and x are positive integers, see http://en.wikipedia.org/wiki/Exponentiation#Failure_of_power_and_logarithm_identities

Improve this answer
$\endgroup$

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.