Wiener function : bug in the doc?

[georgiswan]

Hello,

I was wondering whether there is a bug in the doc of the Wiener filter function:

http://software.intel.com/sites/products/documentation/hpc/ipp/ippi/ippi_ch9/functn_FilterWiener.html

THe doc says that the output is computed as:

Y = u + ( s^2 - n^2 )/ s^2 * ( x - u )

Where:

Y : filtered pixel output

u : average around that pixel

s^2 : stdDev around that pixel

n^2 : estimate of the noise stdDev

x : pixel value (src image)


What happens when s^2 < n^2 ? According to the equation, the output value would be inverted wrt to the input. Is there a sqr missing .. or am I completely lost ?

Thanks!

Gilbert

[PaulF_IntelCorp]

Hello Gilbert,

I don't claim to understand the intricacies of the Wiener function, or its derivation. However; if you peruse the net you'll find this definition is consistent.

Paul

[georgiswan]

Hi Paul,

You are right, I also found the same definition on other sites. I still think that this is bad (possibility of signal inversion -- why would a filter do that?) but it has nothing with the IPP implementation. I probably need to improve my understanding on the basics of the Wiener filter.

Thanks

Gilbert