Applying filters while capturing

[philippas]

Do i lose quality if i apply filters while capturing (tick the RGB filtering option)?
For example, if i capture with virtualdub @ 480x576 YUV2 and use
crop->deinterlace(blendFields)->bicubicResize
Will it be different, quality wise, if i finish the capture and then while compressing again use the above sequence of filters ?

[kempodragon]

You would most likely would lose some quality. Anytime you recompress something that has already been compressed, something is always lost. That's why you should do all filtering and compression after you've captured. Ideally, you should capture full frame with Huffyuv in order to have the maximum amount of detail to work with. Capturing should also be done in YUV, because the computer must convert to RGB, resulting in more loss of quality. Final word: Quality raw material is needed for quality finished product.

[philippas]

I didn't ask if i compress 2 times i will loose quality, that's very obvious!
I asked if i apply filters during capture, will i loose quality ?
Also "enable RGB filtering" works also in YUV2 mode so no colour conversion is done.

With my testing of that feature, i founded a bug in virtualdub.
You cannot change the resolution that the filters are fed, it's always 320x240, which is a bad thing. I founded a way around this, you first choose the target resolution that you want while your not in capture mode, and then enter the capture modewithout closing virtualdub. Then select RGB filtering and the values for the resolution are not flushed and you can use the resolution that you entered before you enter the capture mode.
Correct me if i'm wrong but i believe this is a bug

[kempodragon]

My apologies if I misunderstood. However, I still say you should capture in YUV. The Virtual Dub website states quite clearly that the majority of cards use YUV to capture with. Any RGB output you see has been converted from the original signal. Using YUV will preserve the original signal, as well as lowering the cpu load. Not using filters will lower it even more. I don't know about changing the resolution with filters, since I don't use any filter during capturing (actually I can't, my card doesn't support RGB colorspace :p ). Bottom line, avoid any filtering during capture, save the use of filters for final processing.

[philippas]

@kempodragon
Also "enable RGB filtering" works also in YUV2 mode so no colour conversion is done
My mistake the above is incorect, if you capture in RGB24 or RGB32 there is not colour conversions and you can use the option "Enable RGB Filtering".

The idea is why not choose to do the filtering during capture if your cpu is up to it and save time during the compression of the movie ?

Both times the compressor (i.e xvid,divx) will recieve the frameSize that the resize filter will return(unprocessed frame->Deinterlace filter-->resize filter-->compressor) so i don't see any reason why not apply the filters during capture.

[bb]

There's only one disadvantage of filtering during capture I can imagine of, provided that your CPU is really fast enough to filter in realtime: You don't get the chance to try another filter chain for the same source, because the source is volatile.

I use VirtualDub's built-in denoising during capture, but that's not one of the regular filters you can add in the filters menu. I usually capture to MJPEG, which is sensitive to noise, so I improve the quality of the captured file this way which results in a better quality when re-encoding to MPEG-4 later.


bb

[philippas]

Well with the method that i mensioned i can capture @ 480x576 with huyuv easily since i resize to something like 448x336 during capture, which dramatically reduces the size of the video.
If i don't have a lot of free space capture with xvid @quant-2
with motion search 0 and custom quant Matrix (everything 8).
In my opinion capturing with xvid with the above settings, gives better quality than any Mjpeg that i have tried, since mjepeg introduces block noise and also the filesize is much bigger.

[bb]

You say you resize to 448x336 during capture: how do you deal with combing? When I capture (PAL) the sources are often combed, i.e. phase shifted most of the time. The first filter in the chain I'd need were Telecide, before the resizing, but that costs too much CPU.

If you simply downsize without decombing you'd better drop every second field, i.e. capture at 352x288 or 320x240, respectively. For progressive broadcast your approach sounds reasonable. Unfortunately I never know in advance if the film I am going to capture will be submitted progressive or not...

bb