0
$\begingroup$

I'm reading a book. On pages 105 and 106 it is said that:

For N-look intensity SAR images,
$$I_N=\frac{1}{N}\sum_{i=1}^N I_1(i)=\frac{1}{N}\sum_{i=1}^N (x(i)^2+y(i)^2)$$

Where $x(i)$ and $y(i)$ are the real of the $i$th look (or sample). Since $x(i)$ and $y(i)$ are independently Gaussian distributed, it is well known that $NI_N$ has a Chi-square distribution with $2N$ degrees of freedom. Therefore, the PDF of the N-look intensity is described by $$p_N(I)=\frac{N^N I^{N-1}}{(N-1)!\sigma^{2N}}\exp(-\frac{NI}{\sigma^2}),\quad I\ge 0$$
The mean and the variance are $M_N(I)=\sigma^2$ and $\text{Var}_N(I)=\frac{\sigma^4}{N}$, respectively.

I don't have any problem with $NI_N$ having a chi-square distribution with $2N$ degrees of freedom but as we know the PDF for a chi-square distribution with $f$ degrees of freedom is:
$$h(x) = \begin{cases} \frac{1}{2^{\frac{f}{2}}\Gamma(\frac{f}{2})}x^{\frac{(f-2)}{2}}e^{-\frac{x}{2}}\quad &, x\ge 0\\[2ex] 0\quad &, x\lt 0 \end{cases}$$
so for the distribution of $NI$ we should have:
$$p(NI)=\frac{1}{2^N\Gamma(N)}(NI)^{N-1}\exp(-\frac{NI}{2})=\frac{N^{N-1}I^{N-1}}{2^N(N-1)!}\exp(-\frac{NI}{2})$$
Then if we consider $Y=NI$ as a function of $I$, we have:
$$p(Y)\,dY=p(I)\,dI\;,\;dY=N\,dI\\\frac{N^{N-1}I^{N-1}}{2^N(N-1)!}\exp(-\frac{NI}{2})NdI=p(I)dI\\p(I)=\frac{N^NI^{N-1}}{2^N(N-1)!}\exp(-\frac{NI}{2})$$
Then where does that $\sigma^2$ come from instead of $2$?
As far as I know degree of freedom ($f$) is the only parameter of chi-square distribution. and the mean and the variance of this distribution are $f$ and $2f$ respectively as wiki page says.

$\endgroup$
1
$\begingroup$

I do not know the context, but I would assume that gaussian distribution in the book are not scaled, so their variance is not equal to $1$. Therefore you have another parameter $\sigma^2$.

So what you get is not exactly chi-squared distribution, but chi-squared scaled by a constant. Think of it as Gamma distribution with shape parameter $N$ and scale parameter $\sigma^2/N$.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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