NTSC / 29.XX / Interlaced source problem

[samfchen]

Hello,

I have encountered quite a lot of DVD's of the above mentioned format, and had a hard time dealing with them with the conventional format mentioned by Doom9's Gknot faq. I've done some testing of my own, and following is my general conclusion. I hope everyone who had the same experience can confirm my finding or disprove otherwise.

I've tested a clip of the source (time constraints), with ForcedFilm (Dvd2Avi) and then SmartDeinterlace / Fast Deinterlace under Gknot, Non-forced film, Decomb filter (Telecide / Decimate15), resized, non-resized, and Divx 4.x. Almost all possible REASONABLE combinations of the aforementioned methods were tested.

I find that with the source that I had, the thing that makes the biggest difference (in terms of general clarity, pleasantness of viewing, and few interlacing artifacts while heavy motion) is actually the resizing. I am talking about simply viewing the segment in full screen mode, as opposed to analyzing the segment pixel by pixel. While leaving the bit-rate the same, the resized output seemed to be far inferior then NON-resized output in fullscreen mode. That aside, Decomb seemed to produce a slightly better result then either ForcedFilm + Deinterlace or Forcedfilm alone.

Thank you, and please provide your insights.
Sam:)

[DJ Bobo]

Wai wai wait a moment, you use Forced Film and then use deinterlace???? what's this? a new invention? or the flop of the century?!

You can use Forced Film only on film sources where DVD2AVI shows FILM! and the result will be progressive then, NOT interlaced, so you don't any deinterlacing/IVTC filters!

When you don't have a film source, then don't use Forced Film, save your project, then do IVTC in AviSynth...

Using a lower resolution at the same bitrate gives you always a better result! what are you talking about man?!
You mean 640x480 @ 1000kbps looks better than 448x336 @ 1000kbps? you've gotta be kidding!

You may also consider another thing: ALWAYS do your deinterlacing or IVTC job BEFORE cropping and/or resizing for optimal results!

[pale]

Bobotns wrote:

>>Using a lower resolution at the same bitrate gives you always a >>better result! what are you talking about man?!
>>You mean 640x480 @ 1000kbps looks better than 448x336 @ 1000kbps? >>you've gotta be kidding!

That just is not true. The facts are:
-resizing always has a negative effect on quality (if you resize by an even multiplier such as 2, the decrease in quality is very small though)
-the more bits per pixel you have, the better quality it produces (up to the point of original bitrate, if you go above that, there no gain in quality)

The problem is find the right balance between these two conflicting demands as:
-if you choose not to resize, you have to settle with lower bits/pixel (or more conveniently (bits/(pixel*frame) as used in GKnot)
-if you opt to resize, you suffer from decrease in quality caused by the resizing
-in practise with anamorphic sources, you (pretty much) have to resize vertically

In other words this means to find the resolution where better quality can not be achieved by neither:
-lowering resolution to gain higher bits/(pixel*frame) (and suffereing quality decrease through resizing) nor
-increasing resolution to forego resizing (at least one horizontally)(and suffering from lower bits/(pixel*frame)

With DivX 3.11 the common rule of thumb was that benefits of higher bitrate outweight the disadvantages of resizing with bits/(pixel'frame) below ~0,200.
Some encoders (myself included) have found that with DivX 4.x this sweetspot should be lowered a bit (say to around 0,160-0,170) as DivX 4.x handles low bits/(pixel*frame) better than 3.11 did.

Another approach is to think that you simply want to resize as little as possibly (normally meaning you choose to keep original width of 720, or crop it to 704), and adjust the bitrate to achieve acceptable bits/(pixel*frame). This may (and offen do) call for 2CD movies.