regarding image matching

Hi All,

I have newly joined the group.
I am working on a project which finds defect in a relay,
we r taking images of the relay in all the directions.
we have a source image , to which we have to match template image .
we have used fft to find the correlation .

Problem being faced :
At different Illumination we are having different root mean square
errors, we r doing a grey scale 8 bit image processing.

is there any better algorithms which is illumination independent. Start a New Thread

On Wed, 7 Nov 2007, prathima_rao_rao wrote:

> Problem being faced :
> At different Illumination we are having different root mean square
> errors, we r doing a grey scale 8 bit image processing.
>
> is there any better algorithms which is illumination independent.

Use Normalized Cross Correlation as similarity measure instead of Squared
Difference...

http://www.mlpedia.org/index.php?title=Normalized_cross_correlation

Citation from that page:
"In computer vision, the normalised cross correlation is used as a
distance metric between image patches, as it gives some invariance to
changes in illumination between the patches."

Kind regards,

Filip Rooms

"We all know Linux is great... it does infinite loops in 5 seconds."
- Linus Torvalds about the superiority of Linux on the Amsterdam
Linux Symposium Start a New Thread