Vidcap Display/Capture Settings

[Kevin Culp]

Would anyone be willing to share their experiences optimizing the sharpness setting under VidCap for MJPEG captures with the Rainbow Runner Studio intended for conversion to MPEG-1 or MPEG-2?

Specifically, is it best to use a mid-range setting, use a softer (slightly blurred) image, or go for very crisp? Also, is this setting encoder-dependent?

Thank you,
Kevin

[jeff b]

That's a really good question, Kevin, since the factors involved with two different compression steps can get very tricky.

With MJPEG only, I've always captured with the sharpness slider "slam-dunked" to the left. The reason for my doing it that way is because of the way MJPEG handles tiny, sharp edges, color borders, and noise when the sharpness slider is in any other position. I discovered that my regular 8 captures aren't as prone to this particular problem, but my Hi8 is, due to the higher resolution of the format.

I'm really interested in seeing some more posts to this thread from people who've got a real handle on this, since it's always been a problem for me to get "the most bang for the buck" on MPEG-1 and RealVideo projects. It always seems to be a hit or miss proposition to get these "second step" compressions to look half way decent. Until you posted your question, though, it never occured to me that I might need to start with a sharper MJPEG source capture for these things.

Anyone else have some tips on this?

[ScottV]

Please can somebody tell me exactly where this setting is. I don't remember seeing it anywhere.

[Grigory]

Hi,

I see two questions:
1. What and where is the sharpness settings slider for MJPEG captures, and what it does?
2. What sharpness is good for MPEG1 or 2 production?

I'll try to reply on the second question.

Basically, it is better to reduce the sharpness by applying some smoothing filter. This reduces the high frequency DCT coefficients and helps to escape mosquito artefacts around sharp edges. However, I found that applying strong filtering may not necessary improve overall quality of MPEG movie.
I suppose that applying filter you cause more pixels around moving sharp contoured object to change for each frame or field, not following simple block shift rule. So, the motion extimation has harder time and this may cause blocking artefacts on the scenes with high activity. Blurred frames are somewhat equivalent to noisy frames, because filtering adds "random-like" component to the image.
I tried to check this with bbmpeg encoder at low data rates. It produces very good quality at 5000 kbit/sec with and without say Gaussian blur filter.
If I reduce data rate to 3500, the quality becomes worse, especially on high activity scenes. I can also see mosquito noise around corners.
Introducing Gaussian or similar smoothing does help to remove mosquitoes, but not necessarily reduces blocking. I got several attempts when filtering actually meks the blocking more visible. Another reason is that blocking is more evident on smoothed pictures.

So, I decided that strong filtering does not help to improve MPEG2 quality, or help to reduce the datarate at reasonable quality output.
All above was tested on D8 DV movies. They are quite similar to MJPEG, except less random noise and better sharpness. However, the Digital 8 optical/CCD subsystem has less than 500 lines of resolution, so my original footage is not too sharp. Mosquitoes may be a problem for digital effects, or for SIF size movies because of sharper edges on 8 pixel range. I decided to make only full size MPEG2 files 15-25 min per CD because they preserve motion and look much much better than the best SIF MPEG1 files I could make.

Finally I can only recommend to use smoothing that does not alter perceived sharpness of the picture. Gaussian blur with 0.3 - 0.5 parameter is an example. This helps for DV to Mpeg1 encoding, and does not produce side effects (higher blocking artefacts visibility) on MPEG2.

BTW, the best balance between encoding speed and blocking I found for bbmpeg encoder is to set 12-2 GOP parameter, P frames motion vectors to 20 and 16 Hor / Vert, B frames hor backward and forward vectors to 10, and ver vertors to 8 pixels. Data rate may be in 3450-5700 kbps range.
Increasing the vector length only increase coding time, making them shorter may produce blocking on high activity scenes.

9:25 minutes movie was encoded 18 hours on PII 375 PC.

Grigory

[ScottV]

I only have the first 4 settings. Is this because I'm using a Marvel rather than RR?

[Kevin Culp]

Grigory,

Thanks once again for your detailed response.

When you apply the Gaussian filter to your footage, are you then saving it in an uncompressed (AVI) format before sending it to the bbmpeg encoder? Or are you applying your filters within Premiere and sending it through the bbmpeg plug-in? Since I use MSP 5.2 VE as my editing software, I guess I would be restricted to saving my output to an uncompressed AVI before encoding to MPEG-2.

I guess the question as to what the sharpness setting actually does, and the impact on the final product, is still up in the air... does anyone have any additional thoughts on this issue?

Thank you,
Kevin


[This message has been edited by Kevin Culp (edited 09-22-1999).]

[Patrick]

Scott, in answer to your question, yes, it's possible that you don't have this slider because you have the Marvel. A friend of mine with the Marvel also does not have this slider, but my RR-S does.

It's probably possible to alter the sharpness setting in the Registry if you feel comfortable messing about in there.

[Kevin Culp]

Scott,

If you run the Matrox PC-VCR Remote application, choose "properties", and select the "Video" tab, you'll see five slider bars that let you adjust the brightness, contrast, hue, saturation, and sharpness settings for your video input. If you capture with AVI_IO (as I do), these settings appear under the "Video Settings | Display" menu.

Kevin