|
|
| Morsa |
| Posted: Jan 12 2003, 08:31 PM |
 |
|
Moderator of the Vdub support board
Group: Moderators
Posts: 640
Member No.: 246
Joined: 9-September 02
|
http://forum.doom9.org/showthread.php?s=82...20&pagenumber=1 |
|
| ChristianHJW |
| Posted: Jan 12 2003, 11:05 PM |
|
|
Advanced Member
Group: Moderators
Posts: 1768
Member No.: 2
Joined: 7-July 02
|
Very interesting. Anybody care to explain to me in laymans terms how a wavelet filter is working ?
--------------------
Visit the unofficial Virtualdub support forum on http://forums.virtualdub.org - help to reduce the big number of emails Avery Lee is getting every day !!
Support matroska as container and Gstreamer as the only truely open, x-platform multimedia platform .... |
|
| Morsa |
| Posted: Jan 13 2003, 05:11 AM |
|
|
Moderator of the Vdub support board
Group: Moderators
Posts: 640
Member No.: 246
Joined: 9-September 02
|
This is a filter to remove some kind of noise that cannot be removed with atempral filter.
It is based on wavelet theory, like the wavelet codecs for video.I think (although I'm not sure) that this has something to be with Frequency domain instead of spatial like Deconvolution.
I've read something about deconvolving using wavelets too. |
|
| Sarreq Teryx |
| Posted: Jan 13 2003, 05:18 AM |
|
|
VirtualdubMod Alpha tester
Group: Vdubmod Alpha Testing Team
Posts: 175
Member No.: 41
Joined: 16-July 02
|
Anyone happen to know if there is a non-sse version floating about
--------------------
And as I walk through the Valley of the Shadow of Death, Lord, thy balls and shaft, they comfort me, you annoint my head with oil, some salt, a dash of pepper, a sprigg of parsley......Lord?............Lord??? What dost thou intend to do with that fork??? |
|
| phaeron |
| Posted: Jan 13 2003, 08:50 AM |
|
|
Virtualdub Developer
Group: Administrator
Posts: 7773
Member No.: 61
Joined: 30-July 02
|
The discrete wavelet transform (DWT) decomposes a signal into two half-frequency components: a low-frequency signal, and a high-frequency signal. Probably the best way to imagine it is to "peel" off the details from an image, and then squish both the detail and the soft image to half size. Wavelets were popular in video compression some years back around the time of the Indeo codec -- the low frequency component is repeatedly passed through the DWT, forming a multiresolution pyramid which is then quantized and entropy encoded. The DWT is faster to compute than the DCT (O(n) vs. O(n log n)), can be computed exactly with an appropriately chosen kernel, and has a wonderful scalability property: since decompression is recursive, you can decode any power-of-two scaled version of the source signal by simply throwing out the higher resolution sample sets, something like the way mipmaps work. Wavelet transforms also have better temporal locality than DCTs -- a disturbance in wavelet space only causes a small disturbance in the signal, something like a bit of ringing, but a disturbance in DCT space changes the entire signal (i.e. you get crap throughout the entire block).
As I understand it, since the DWT is nothing more than convolution by two FIR kernels followed by decimation -- and the second kernel is merely the spectral reverse of the first -- there is nothing magical about the DWT that cannot be performed with regular convolution (FIR) or FFT methods. It's simply a convenient way of slicing the spectrum of a signal into "detail" and "non-detail" parts, which can then be conveniently handled for compression and noise reduction purposes. Presumably, the wavelet filter uses the DWT to quickly separate out the low-frequency parts of the image and keep them away from the noise reduction algorithm.
I ain't no electrical engineer, however, so anyone who has actual training in signal processing is free to jump in and correct me.
|
|
