Why xvid looks better than mpeg2?

[computerchuck]

Here is what I did. I captured music video of pussycat dolls beep off of Dishnetwork MTV. Capture card is WinTV PVR2 USB2.0. MPG2 DVD Quality CBR 6400kps. Picture looks ok. Just for the heck of it, I loaded it into autogk and did a xvid conversion. It doubled the file size, but it also made the video look better. Here is where I am confused. If the original file didn't look that good, what did it do to clean up the image? It looks blurry on mpg playback but crisp and sharp on the xvid playback minus the distortions on fast movements on the dancers. Is there a way to improve the mpg picture without using xvid? basically a filter to pass it through?

[mats.hogberg]

Not with WinTV itself, but you can do the same kind of filtering reencoding to mpeg2 as to XviD. In theory, you can never gain quality by reencoding a video, but by applying the right filters, you can make it look better.


/Mats

[guns1inger]

Did it keep the original resolution, or reduce it down ? Resizing down can hide a lot of problems

[mats.hogberg]

XviD @ 12800 kbps is really something, as the "std" DivX/XviD bitrate used is generally around 1/10 of that.
Also the distortion in fast movement can be interlace lines - on a TV it will look as it should.

/Mats

[manono]

If you can't decipher the AutoGK log to figure out what it did, you could always post it here. It may have IVTC'd it or deinterlaced it, which could easily have made it look better than viewing the MPEG-2 on the computer monitor through some software players.

Sure, anything AutoGK did can be duplicated on the MPEG-2 using the same filters. It might not be necessary though, if your final viewing is to be the TV set. The problems you saw when viewing the MPEG-2 on the computer monitor could easily disappear when viewing on a standard interlaced TV set.

[SingSing]

There are break thru, that breaks the old rule that said the best looking video has to be the original.

The new PowerDVD 6 has CLEV and CLPV and smart deinterlace, all these together make the video looks brighter and cleaner on Computer.

[jagabo]

Is there a way to improve the mpg picture without using xvid?

You can adjust the PVR USB2's sharpness and spacial/temporal smoothing using tools from http://www.shspvr.com .

[thecoalman]

the old rule that said the best looking video has to be the original.
.

Filter and small adjustments aside.... it always will be. No matter how well the filter can enlarge, de-noise or whatevr adjustment it makes it's going to lack the one fundamental thing it needs to really improve and that's detail. There's no magic bullet for that and never will be, you can't improve what's not there.

[SingSing]

Eyes can be fooled. The most common example is we cannot see the flickering of any 50Hz or 60Hz light bulbs.

There are products that up-sample digital photo, and Cyberlink is trying to improve the look of video on PC.

[SatStorm]

nvidia's pure video also...

[thecoalman]

There are products that up-sample digital photo,

Yes and they do a great job of filling in areas of solid color and preserve the edges. But in the end all you have is giant featurless soft image. they work fantastic on images like cartoons but there's no detail in a cartoon image.

and Cyberlink is trying to improve the look of video on PC.

Power DVD does a great job.... But that's irrelevant since it does a great job of reproducing what you already have on TV, it doesn't improve anything except the computer playback.

[vitualis]

When you converted it to XviD, it probably smoothed out some of the noise in the original, hence looking "better". Detail would have been lost though.

Regards.

[SingSing]

There are products that up-sample digital photo,

Yes and they do a great job of filling in areas of solid color and preserve the edges. But in the end all you have is giant featurless soft image. they work fantastic on images like cartoons but there's no detail in a cartoon image.

Upsampling digital Photo is not filled in the blank. They use tool like Fracture Math. You under estimate the morden Science and Math.

[thecoalman]

Upsampling digital Photo is not filled in the blank. They use tool like Fracture Math. You under estimate the morden Science and Math.

Do you mean fractal math... I'm well aware of the techniques they use having worked with digital images extensively over the last 10 years. Plain and simple you cannot create what is not there. You get a soft blurry image.. see my avatar to the left to compare.

[guns1inger]

I think you over estimate modern science and maths. You can only do this stuff on CSI, and any lab tech can do it in three lines of code.

[vitualis]

There are some good upsampling algorithms, but as others, you can't create detail that isn't there.

The good algorithms try to detect edges so that after upsampling they are still relatively sharp (as compared to, say, a standard bicubic or bilinear resize).

Regards.

[jimmalenko]

I think you over estimate modern science and maths. You can only do this stuff on CSI, and any lab tech can do it in three lines of code.

... and a 3 month modelling CGI render takes 5 seconds to "build"

[thecoalman]

(as compared to, say, a standard bicubic or bilinear resize).

I used bicubic for the one above, I'm sure there are other resizers capable of preserving the edges better but it's not going to be that much. i was recently looking at resizing filter specifically for sampling up. They had some examples there and the source images and the results... so of course I compared to what I have now. The sample image with a lot of detail wasn't much better than using bicubic. They also had a cartoon image and that is where it excelled...the edges were perfect. Much better than what I could produce.

[SatStorm]

Upscaling is something that always bothered me with mpeg sources.
Most European DVB channels are 544 x 576, so I have to choose to convert to 352 or 704 x 576.
There are now some great looking DVB channels, that indeed loose something when you convert to 352x576. Bicubic is the best way to upscale the 544 to 704 horizontal lines, but which method is the best I don' t really know! What do you use for this (with virtualdub)? I would love to read your opinions!

[vitualis]

(as compared to, say, a standard bicubic or bilinear resize).

I used bicubic for the one above, I'm sure there are other resizers capable of preserving the edges better but it's not going to be that much. i was recently looking at resizing filter specifically for sampling up. They had some examples there and the source images and the results... so of course I compared to what I have now. The sample image with a lot of detail wasn't much better than using bicubic. They also had a cartoon image and that is where it excelled...the edges were perfect. Much better than what I could produce.

The most well known product is probably Genuine Fractals: http://www.imaging-resource.com/SOFT/GF/GF.HTM

It actually DOES work pretty well.

However, whether it is much better than pre-filtering the image (with something like Neat Image) and then doing "step" upscaling (e.g., using standard bicubic but only upsizing by 10% at a time until you reach your target size) is debatable.

Regards.

[jagabo]

The most well known product is probably Genuine Fractals: http://www.imaging-resource.com/SOFT/GF/GF.HTM

It actually DOES work pretty well.

Search for other reviews of Genuine Fractals. Most are not as friendly as that one.

[edDV]

I've heard Genuine Fractals is good for enhancing video still captures to poster size. I've not heard much about fractals being used for general video upscale.

Video doesn't perform well with that much frame to frame randomness unless it is totally decompressed. Compression software relies on frame to frame redundancy.

[thecoalman]

The most well known product is probably Genuine Fractals: http://www.imaging-resource.com/SOFT/GF/GF.HTM

It actually DOES work pretty well.


Regards.

Can you do a sample from my avatar?

[vitualis]

I don't use it. It doesn't do all too much and it's expensive.

If I need to "upsize" (and want best possible quality) then I use what I stated before. I may or may not filter with Neat Image (depending on source), and then use step-wise upsampling by 10% increments. The only time I really do this is if I'm upsampling a digital photo before taking it for developing. If I need a say 15"x10" image at 300 dpi, I prefer to prepare in exactly that format rather than let some automated gizmo at the photo developers do it for me.

You can find demos of it on the net. It does work reasonably well though once you blow up an image to ridiculously large sizes from the original, you get a funny smoothed out cartoon look. As before, you can't generate detail out of nothing, but it still looks better than than a straight single upsize using bicubic.

Regards.

[vitualis]

Image examples:
http://graphicssoft.about.com/od/resolution/a/genuinefractals.htm
http://www.kenrockwell.com/tech/gf.htm

Examples including using stepwise resizing:
http://www.camerahobby.com/Digital_GenuineFractals_Stairstep.htm

IMHO, GF doesn't offer me much above stepwise resizing with bicubic... and I can do that for free. Considering that I do this maybe several times a year at most, GF is pretty pricey.

Regards.

[thecoalman]

It does work reasonably well though once you blow up an image to ridiculously large sizes from the original, you get a funny smoothed out cartoon look. .

http://graphicssoft.about.com/library/extra/ngf2r-test2.htm

You can see that here, it almost looks like one of those "oil painting" filters was applied.

[mats.hogberg]

Folks, aren't we on the verge of threadjacking here?

/Mats

[computerchuck]

what software program removes video noise from an mpeg source hopefully without re-encoding if possible?

[mats.hogberg]

what software program removes video noise from an mpeg source hopefully without re-encoding if possible?

Unforunately, there's no way of applying any kind of filter (including video noise filter) without reencoding.
I'd use AviSynth for filtering and frameserving to the encoder.

/Mats