| Steven Pigeon 2007-02-24, 6:55 pm |
| K. wrote:
> what is the connection between the two procedures in terms that PCA
> finds the directions of most variance and I preselected the
> coefficients with largest variances? shouldn't PCA do the same thing?
> is the correct conclusion that the rest of the coefficients (the
> remaining ~15000) are simply redundancy in terms of face recognition?
The PCA is like most transforms: it does "energy packing". The largest
coefficients correspond to largest variances, to large-scale structures
of the image. The smallest coefficients, of very low amplitude,
correspond to small-scale structures of the image. Low amplitude+small-
scale usually means 'noise', like the fine gain random noise in a tv
picture. Stripping low amplitude/small scale details is basically a
low-pass noise filtering.
Low pass filtering / noise removal will of course help recognition since
now you compare only large-scale structures rather than random pixel
fluctuations.
Best,
S.
> the main question: how does preselecting the coefficients with highest
> variance across all images correlate to what standard PCA does? can
> this be compared at all since one original space is 512-D and the
> other 16384-D?
>
> thanks for you help.
> K.
>
|