Software filters in a nutshell, denoising and deinterlacing VHS video captures

[guokamoli]

At the moment, I dump VHS and Betamax videos (PAL) through an AVT-8710, through a Canopus ADVC110, through the program WinDV, onto my computer. The footage is then loaded up through virtualdub, and modified as such.

- deinterlaced (double frame rate)
- downscaled from 720x288 to 'approx' 352 x 'exactly' 288 (Lancoz 3)
- De-noised using "Neat video" filter
- Upscaled twice using "Deemon Video Enhancer" (from '352'x288 to '704'x576, then again to '1408'x1152)
- Borders are carefully cropped (with assistance from a ratio calculator to maintain a 5:4 ratio).
- Footage is downscaled to 720x576
- Footage is re-interlaced
- Footage is saved as uncompressed avi for later conversion with TMPGenc

At the moment, I am yet to research a 'better' interlacer that can interlace a 50fps 720x288 media to 25fps 720x576 (will make the downscaling in the previous step more effective).

I have also experimented with Colormill...but I hate it. Its good for optimising colour in a single scene, but I find it useless when converting a entire video, as there is no universal setting that will work for a video with millions of different scenes.

All in all, for every hour of footage, it takes approximately 24 hours to filter the footage using the above filters. This leads me to think that I will either need more computers to make the processing timeframe more practical, or...maybe I'm just not doing this right

Am I on the right track?

[kpmedia]

Quote:

- deinterlaced (double frame rate)
- downscaled from 720x288 to 'approx' 352 x 'exactly' 288 (Lancoz 3)
- De-noised using "Neat video" filter
- Upscaled twice using "Deemon Video Enhancer" (from '352'x288 to '704'x576, then again to '1408'x1152)
- Borders are carefully cropped (with assistance from a ratio calculator to maintain a 5:4 ratio).
- Footage is downscaled to 720x576
- Footage is re-interlaced

I've looked at this for a few days now, and want to try it myself (not yet had time), but didn't want this to sit much longer without a reply. Honestly, that chain of filters makes me squirm a bit, because I'm not sure if all the scaling is doing more harm than good. Generally speaking, you want to re-scale footage as few times as possible, because you'll start to exchange actual detail and sharpness for noise and artifacts. Downscaling will fuzz out detail, while upscaling will sharpen both noise and image.

Deinterlacing, even side-by-side field separations, don't always work as planned. I know that all of the VDub side-by-side deinterlacers are imperfect, and can leave motion artifacts after certain filters are applied (between de-interlace and re-interlace). Median is one such filter. There are several more, but I don't remember them off the top of my head (which also means they're not regular-use type filters).

Quote:

- Footage is saved as uncompressed avi for later conversion with TMPGenc

This does work, and in some situations it works well. But it's also just a $37 program that is more than a decade old. That does NOT mean it's bad, but it does mean there are some better encoders that now exist. MainConcept Reference, for example ($510 USD for the MPEG-2 codec). If you're going to be doing professional transfers, you really should swap to the MC based encoding platform, because it's much cleaner and makes much better use of bit rates.

Quote:

I have also experimented with Colormill...but I hate it. Its good for optimising colour in a single scene, but I find it useless when converting a entire video, as there is no universal setting that will work for a video with millions of different scenes.

There's not other way to color correct, outside of scene-by-scene work. There's no automatic "fix color" type software out there. (Well, there probably is, but the seller is basically lying. It's just not possible to do.) An editor like Adobe Premiere Pro, and possibly Adobe Premiere Elements, makes it easier to color correct, because you can knife the scenes on the timeline, effectively creating separate clips without need of actually manual creation of clips (as would be required using VirtualDub as an editor).

ColorMill is very good, but Premiere lends itself to an easier workflow. No disagreements there!

Quote:

All in all, for every hour of footage, it takes approximately 24 hours to filter the footage using the above filters.

The more filters you add, the longer it takes. It's yet another side effect of possibly using too many filters. NeatVideo is also a resource hog (definitely worth it -- but still a great pig pig on the CPU and RAM).

Quote:

This leads me to think that I will either need more computers to make the processing timeframe more practical

Any serious enthusiast, hobbyist or professional pretty much requires use of multiple computers. Some people get bent out of shape, thinking they'll have a room full of monitors, desks, etc -- but that's not the case. Get a nice 4-port KVM, and you can use the same monitor, mouse, keyboard and set of speakers for all four systems at the same time. You can't use more than one computer anyway, so you only need input/output devices when it's the system getting some attention. TrendNet KVMs are excellent, and very inexpensive (~$50 for a 4-port KVM, ~$25 for 2-port KVM). Get the "with audio" models. Airlink also used to make a great KVM, almost identical to the TrendNet. These days, you'll probably only find used Airlink KVMs for about the same price as the new TrendNets.

Quote:

Am I on the right track?

More or less, yes.

[guokamoli]

Quote:

Originally Posted by kpmedia

Generally speaking, you want to re-scale footage as few times as possible, because you'll start to exchange actual detail and sharpness for noise and artifacts.

hmm...so perhaps I should invest in another analogue>didgital conversion device that can save footage at a more native resolution then, as opposed to an initial downscale?

Quote:

Originally Posted by kpmedia

Downscaling will fuzz out detail, while upscaling will sharpen both noise and image.

Yeah, youre not wrong there. At the moment, I am using lancoz3 methods to downscale (I felt it would be more efficient than nearest neighbour), and Deemon Video Enhancer plugin for vdub to upscale. Are these the best downscalers/upscalers at this time?

On a slightly related note, would that $8000 ikena reveal software upscale better than video enhancer, or is it all a bunch of false hype?

Quote:

Originally Posted by kpmedia

Deinterlacing, even side-by-side field separations, don't always work as planned. I know that all of the VDub side-by-side deinterlacers are imperfect, and can leave motion artifacts after certain filters are applied (between de-interlace and re-interlace).

I guess I should do further research into a solid interlacer 'and' deinterlacer then

Quote:

Originally Posted by kpmedia

MainConcept Reference

I'm willing to part with $510 for this product, but is this product definitely the way to go for encoding at this time? Would there be any others that do a finer job?

Quote:

Originally Posted by kpmedia

There's not other way to color correct, outside of scene-by-scene work. There's no automatic "fix color" type software out there.

Well... I have been experimenting with colour enhancers actually. The means are impractical, but I have been getting some nice 'automatic' results when I convert a video to an image sequence, and then use batch photo editors to automatically improve colour, brightness, and white levels. Then I just convert the image sequence back into an uncompressed avi. Takes shitloads of cpu time though

...can I say shitloads?

Quote:

Originally Posted by kpmedia

Any serious enthusiast, hobbyist or professional pretty much requires use of multiple computers.

Sounds like a plan then...but I might wait until the end of the financial year before I hopefully claim a tax refund for this project

On a simular note, what would be the ideal budget for a 'number' of video conversion rigs? Maybe $1000 each?

Quote:

Originally Posted by kpmedia

More or less, yes.

Awesome. Thanks kpmedia

[kpmedia]

Quote:

Would there be any others that do a finer job?

Finer than MainConcept Reference? Not a chance. Not for MPEG-2, not for H.264. You have Mainconcept, and then you have everything else. Everybody else copies MainConcept, or at least tries to copy them. MC has set the bar for encoding quality very, very high!

Quote:

...can I say shitloads?

Well .... I think "loads" would suffice on its own merit.

Quote:

On a simular note, what would be the ideal budget for a 'number' of video conversion rigs? Maybe $1000 each?

Define what would entail a "rig" if you would. List all hardware and/or software. $1K is a good budget, but it depends on what all you're trying to buy with it. Inversely, there's no need to throw away money on something meaningless. CPUs are a good money pit, for example, if you're a sucker for buying the latest one that comes out.

[guokamoli]

Quote:

Originally Posted by kpmedia

Finer than MainConcept Reference? Not a chance.

So Mainconcept reference is the Chuck Norris of MPEG-2 'and' H.264 encoding? Gotcha

Quote:

Originally Posted by kpmedia

Well .... I think "loads" would suffice on its own merit.

But then where would be the fun in that?

Quote:

Originally Posted by kpmedia

Define what would entail a "rig" if you would.

Alright, letsee:

CPU: Intel Core i7 2600k 3.4Ghz ($319AU)
MOBO: Gigabyte GA-Z68MA-D2HB3 (w onboard video) ($159AU)
RAM: 8GB DDR3 1333 Kingston Memory (2x 4GB) ($159AU)
CASE: Coolermaster Centurion 5 II (w 500W PSU) ($108AU)
HDD: 3Tb WD30EZRS/X ($178AU)
DISC DRIVE: LG BH10LS30 Blu Ray Writer ($99AU)

All up, the hardware weighs in at $1022AU from netplus.com.au. Obviously the software will cost extra

Considering that I may be investing in multiple towers, would a rig like the one above be the way to go?

[admin]

Right way to go? For capture systems, no. Although some of that would depend on the cards in use.

For encoding systems, yes, that would work.

The biggest mistake people make is throwing huge sums of money at CPUs, when it's really not as important as the common myth/mantra makes it out to be. Yes, a fast CPU is good. But any decently fast CPU suffices. When it comes to video work, look at the core speed more than anything else. At 3.4Ghz per core, you're doing well. But still keep in mind that quite a bit of video encoding software only sees 1-2 cores. Even those that see more cores cannot actually use it because of bottlenecks in the storage (hard drives). And no, throwing RAID and faster drives at it doesn't necessary help. Patience is really the ultimate skill for video. Few people have it.

8GB of RAM doesn't do anything for you. Not for video. The only exception may be Adobe Premiere, when using complex timelines. A program like Photoshop uses RAM more than any video tool will need.

I would get Pioneer Blu-ray burners, not LG.

When it comes to cases and power supplies, go for anything that gives you low noise. Otherwise it will be impossible to do audio work with a room full of airplanes creating a wind tunnel our of your workspace.

[guokamoli]

Quote:

Originally Posted by admin

For encoding systems, yes, that would work.

Its an encoding system

Quote:

Originally Posted by admin

The biggest mistake people make is throwing huge sums of money at CPUs, when it's really not as important as the common myth/mantra makes it out to be.

Maybe...but then I use a ridiculous amount of virtualdub plugins simultaneously. On my Toshiba Qosmio X500, it appears that the bottlekneck during video processing lies in the CPU. Everything else keeps up comfortably, including the 2TB Seagate Green Barracuda that the uncompressed video is written to

All 8 Virtual cores are being consumed by Virtualdub.

Quote:

Originally Posted by admin

8GB of RAM doesn't do anything for you...a program like Photoshop uses RAM more than any video tool will need.

Perhaps...but it will come in handy if I wanted to trial further experiments (ie. batch photoshop commands to enhance colour from image sequences 0 )

[Steve(MS)]

If and that is the word...if you could put a hardware color corrector, like a good proc amp, it may
make your work easier once the file gets on hd.
I found out early on how right Admin and LS are about hardware correction vs software.
Even if the proc doesn't cure all your problems, hopefully the following corrections will be more manageable.
Helps too to have a highend vcr to start with.
Does DV change colors?

[admin]

Quote:

Does DV change colors?

It can, yes. In fact, it often does with certain pieces of hardware. It's more pronounced with NTSC analog source video, which is ripped down from 4:2:2 comparable colorspace to the DV 4:1:1 colorspace. PAL isn't quite so harsh, dropping down to 4:2:0 (which yields more color data than 4:1:1). According to theory and die-hard DVD fans, it's "good enough" -- a term generally used to downplay or ignore obvious flaws.

DV was created as a shooting format, not a video conversion format. DV-shot video doesn't really have these issues. It's the converted videos that have that digitally cooked look.

To be fair, however, most hardware alters colors (chroma, luma, IRE) by some degree, when analog video is converted into digital domain video. This process can be further compounded by the user's choice of codecs and software. This is why it's so important to use known-quality hardware and capture software. Carefully select hardware, codecs and software, when capturing video -- as well as carefully selecting everything that comes before it in the wokflow (i.e., cameras, VCR/VTR, TBCs, added video/audio hardware, etc). You can mitigate and prevent certain damage from happening by simply using intelligent transfer methods.