Hi Nicolas, Nicolas schrieb:
> I don't apply any filter. I spent 2 evenings trying to find good > >settings to denoise the video without any success. Each time the result >was blur. There was far less details on the pictures... > > hmm,... you could (do you use the CVS version of the denoiser?) use -M 0,0,0 -m 0,0,0 -t 8,16,16... Blurring only should be visable when using too high thresholds for the spatial filters (-M and -m). When appropriatly set you shouldn't be able to see any blurring even with these. The above example, however, turns the spatial filters off, completely. The temporal filter, however, can *not* reduce sharpness (it might, if using thresholds above 24 (which is a really huge setting!) cause visable ghosts.) So, if you want to denoise without loosing sharpness by all means use only -t... BTW: if it is any analoge video tape, you should use horizontal lowpass-filtering, at least *once* in your chain. Here comes why (you may not like to hear... Warning, long...): You say it's a recording of a HI8 tape. These have a bandlimit of 3Mhz (luminance) applied before recording the video signal on tape. Sampling theory says you'll be needing 6.000.000 samples/sec to get a fully reconstructable digital recording of such a signal. Samples is pixels for digital images. Asuming you're recording a PAL signal to such a tape you have 25 frames (= 2 fields) per second. This makes 6.000.000/25=240.000 pixels per frame. PAL has 625 scanlines per frame, so you get 240.000/625=384 pixels/scanline. This is further reduced by the horizontal blank-interval. For PAL it has a time-period of 4,7µs. The complete scanline has a time-period of approximatly (I ignore vsync here...) 64µs. So approximatly 7,3% of the complete scanline is lost for image transmission. This makes 384*(1-0,073)=356 pixels of resolution. Considering the active lines of a PAL-video (and I don't know a digital-video-system which stores more than that) to be 576, you will end up with an image of 356x576 active/usable pixels by maximum. For a standard (that is luma correctly bandlimited to 5Mhz) PAL-signal you will end up with an image of 593x576 pixels. Everything you record above these limits is *noise*... That noise however may mask how bad the HI8 recording really is... Furthermore the maximum vertical resolution of an interlaced image is reduced by the Kell-factor to approximatly 2/3*nr_of_lines. This is reflected in the method any good interlaced TV/video-camera will sample the image off the CCD. Every halfway sane manufacturer will use "double-scanline-readout" to 1. reduce noise (+3dB SNR) 2. reduce inter-field-flicker 3. reduce inter-field jumpyness of scanlines 4. honour the Kell-factor as otherwise fine detail may not recorded at all (if it has the correct, critical velocity) That is, given this is the CCD: 1. line 2. line 3. line 4. line 5. line . . . For the first field (bottom field first) scanlines (2+3)/2 and (5+4)/2 and so on are read out, for the second field it's (1+2)/2, (3+4)/2, ... and so on... Everyone clearly can see this will reduce resolution before recording anything. To get that lovely crisp images we all like from (cheap) video-cameras, boosting of mid to high frequencies is applied, which makes the image even more worse... >minutes long Hi8 cassette > >I fear that could reduce the video quality. > It's HI8... What quality? ;-) (Sorry, could not resist...) cu Stefan -- Gnomemeeting/Netmeeting: callto:ils.seconix.com/[EMAIL PROTECTED] ICQ: 131490319 ------------------------------------------------------- This SF.Net email is sponsored by xPML, a groundbreaking scripting language that extends applications into web and mobile media. Attend the live webcast and join the prime developer group breaking into this new coding territory! http://sel.as-us.falkag.net/sel?cmd=lnk&kid0944&bid$1720&dat1642 _______________________________________________ Mjpeg-users mailing list Mjpeg-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/mjpeg-users