MPlayer - The Movie Player: http://www.mplayerhq.hu | ||
---|---|---|
Előző | Fejezet 7. Encoding with MEncoder | Következő |
Introduction. I suggest you visit this page if you don't understand much of what is written in this document: http://www.divx.com/support/guides/guide.php?gid=10 This URL links to an understandable and reasonably comprehensive description of what telecine is.
For technical reasons pertaining to the limitations of early television hardware, all video intended to be displayed on an NTSC television set must be 59.94 fields per second. Made-for-TV movies and shows are often filmed directly at 59.94 fields per second, but the majority of cinema is filmed at 24 or 23.976 frames per second. When cinematic movie DVDs are mastered, the video is then converted for television using a process called telecine.
On a DVD, the video is never actually stored as 59.94 fields per second. For video that was originally 59.94, each pair of fields is combined to form a frame, resulting in 29.97 frames per second. Hardware DVD players then read a flag embedded in the video stream to determine whether the odd- or even-numbered lines should form the first field.
Usually, 23.976 frames per second content stays as it is when encoded for a DVD, and the DVD player must perform telecining on-the-fly. Sometimes, however, the video is telecined before being stored on the DVD; even though it was originally 23.976 frames per second, it becomes 59.94 fields per second, and is stored on the disk as 29.97 frames per second.
When looking at individual frames formed from 59.94 fields per second video, telecined or otherwise, interlacing is clearly visible wherever there is any motion, because one field (say, the even-numbered lines) represents a moment in time 1/59.94th of a second later than the other. Playing interlaced video on a computer looks ugly both because the monitor is higher resolution and because the video is shown frame-after-frame instead of field-after-field.
Notes:
This section only applies to NTSC DVDs, and not PAL.
The example MEncoder lines throughout the document are not intended for actual use. They are simply the bare minimum required to encode the pertaining video category. How to make good DVD rips or fine-tune libavcodec for maximum quality is not within the scope of this document.
There are a couple footnotes specific to this guide, linked like this: [1]
Progressive video was originally filmed at 23.976 fps, and stored on the DVD without alteration.
When you play a progressive DVD in MPlayer, MPlayer will print the following line as soon as the movie begins to play:
demux_mpg: 24fps progressive NTSC content detected, switching framerate.From this point forward, demux_mpg should never say it finds "30fps NTSC content."
When you watch progressive video, you should never see any interlacing. Beware, however, because sometimes there is a tiny bit of telecine mixed in, where you wouldn't expect. I've encountered TV show DVDs that have one second of telecine at every scene change, or at seemingly random places. I once watched a DVD that had a progressive first half, and the second half was telecined. If you want to be really thorough, you can scan the entire movie:
mplayer dvd://1 -nosound -vo null -benchmarkUsing
-benchmark
makes
MPlayer play the movie as quickly as it
possibly can; still, depending on your hardware, it can take a
while. Every time demux_mpg reports a framerate change, the line
immediately above will show you the time at which the change
occurred.
Sometimes progressive video is referred to as "soft-telecine" because it is intended to be telecined by the DVD player.
Telecined video was originally filmed at 23.976, but was telecined before it was written to the DVD.
MPlayer does not (ever) report any framerate changes when it plays telecined video.
Watching a telecined video, you will see interlacing artifacts that seem to "blink": they repeatedly appear and disappear. You can look closely at this by
mplayer dvd://1 -speed 0.1
Seek to a part with motion.
Look at the pattern of interlaced-looking and progressive-looking frames. If the pattern you see is PPPII,PPPII,PPPII,... then the video is telecined. If you see some other pattern, then the video may have been telecined using some non-standard method and MEncoder cannot losslessly convert it to progressive. If you don't see any pattern at all, then it is most likely interlaced.
Sometimes telecined video is referred to as "hard-telecine".
Interlaced video was originally filmed at 59.94 fields per second, and stored on the DVD as 29.97 frames per second. The interlacing is a result of combining pairs of fields into frames, because within each frame, each field is 1/59.94 seconds apart.
As with telecined video, MPlayer should not ever report any framerate changes when playing interlaced content.
When you view an interlaced video closely with -speed 0.1
,
you will see that every single frame is interlaced.
All of a "mixed progressive and telecine" video was originally 23.976 frames per second, but some parts of it ended up being telecined.
When MPlayer plays this category, it will (often repeatedly) switch back and forth between "30fps NTSC" and "24fps progressive NTSC". Watch the bottom of MPlayer's output to see these messages.
You should check the "30fps NTSC" sections to make sure they are actually telecine, and not just interlaced.
In "mixed progressive and interlaced" content, progressive and interlaced video have been spliced together.
This category looks just like "mixed progressive and telecine", until you examine the 30fps sections and see that they don't have the telecine pattern.
As I mentioned in the beginning, example MEncoder lines below are not meant to actually be used; they only demonstrate the minimum parameters to properly encode each category.
Progressive video requires no special filtering to encode. The only
parameter you need to be sure to use is
-ofps 23.976
. Otherwise, MEncoder
will try to encode at 29.97 fps and duplicate frames.
mencoder dvd://1 -nosound -ovc lavc -ofps 23.976
Telecine can be reversed to retrieve the original 23.976 content,
using a process called inverse-telecine.
MPlayer contains two filters to
accomplish this: detc
and
ivtc
. You can read the manual page to see their
differences, but for DVDs I've never had a problem with
ivtc
. Note that you should
always inverse-telecine before any
rescaling; unless you really know what you're doing,
inverse-telecine before cropping, too
[1]. Again,
-ofps 23.976
is needed, too.
mencoder dvd://1 -nosound -vf ivtc=1 -ovc lavc -ofps 23.976
For most practical cases it is not possible to retrieve a complete progressive video from interlaced content. The only way to do so without losing half of the vertical resolution is to double the framerate and try to "guess" what ought to make up the corresponding lines for each field (this has drawbacks - see method 3).
Encode the video in interlaced form. Normally, interlacing wreaks
havoc with the encoder's ability to compress well, but
libavcodec has two
parameters specifically for dealing with storing interlaced video a
bit better: ildct
and ilme
. Also,
using mbd=2
is strongly recommended
[2] because it
will encode macroblocks as non-interlaced in places where there is
no motion. Note that -ofps
is NOT needed here.
mencoder dvd://1 -nosound -ovc lavc -lavcopts ildct:ilme:mbd=2
Use a deinterlacing filter before encoding. There are several of
these filters available to choose from, each with its own advantages
and disadvantages. Consult mplayer -pphelp
to see
what's available (grep for "deint"), and search the
MPlayer mailing lists to find many discussions about the
various filters. Again, the framerate is not changing, so no
-ofps
. Also, deinterlacing should be done after
cropping [1] and
before scaling.
mencoder dvd://1 -nosound -vf pp=lb -ovc lavc
Unfortunately, this option is buggy with
MEncoder; it ought to work well with
MEncoder G2, but that isn't here yet. You
might experience crahes. Anyway, the purpose of -vf
tfields
is to create a full frame out of each field, which
makes the framerate 59.94. The advantage of this approach is that no
data is ever lost; however, since each frame comes from only one
field, the missing lines have to be interpolated somehow. There are
no very good methods of generating the missing data, and so the
result will look a bit similar to when using some deinterlacing
filters. Generating the missing lines creates other issues, as well,
simply because the amount of data doubles. So, higher encoding
bitrates are required to maintain quality, and more CPU power is
used for both encoding and decoding. tfields has several different
options for how to create the missing lines of each frame. If you
use this method, then Reference the manual, and chose whichever
option looks best for your material. Note that when using
tfields
you
have to specify both
-fps
and -ofps
to be twice the
framerate of your original source.
mencoder dvd://1 -nosound -vf tfields=2 -ovc lavc -fps 59.94 -ofps 59.94
If you plan on downscaling dramatically, you can excise and encode
only one of the two fields. Of course, you'll lose half the vertical
resolution, but if you plan on downscaling to at most 1/2 of the
original, the loss won't matter much. The result will be a
progressive 29.97 frames per second file. The procedure is to use
-vf field
, then crop
[1] and scale
appropriately. Remember that you'll have to adjust the scale to
compensate for the vertical resolution being halved.
mencoder dvd://1 -nosound -vf field=0 -ovc lavc
In order to turn mixed progressive and telecine video into entirely progressive video, the telecined parts have to be inverse-telecined. There are two filters that accomplish this natively, but a better solution most of the time is to use two filters in conjunction (read onward for more detail).
Currently the most reliable method to deal with this type of video
is to, rather than inverse-telecine the telecined parts, telecine
the non-telecined parts and then inverse-telecine the whole
video. Sound confusing? softpulldown is a filter that goes through
a video and makes the entire file telecined. If we follow
softpulldown with either detc
or
ivtc
, the final result will be entirely
progressive. Cropping and scaling should be done after the
inverse-telecine operations, and -ofps 23.976
is
needed.
mencoder dvd://1 -nosound -vf softpulldown,ivtc=1 -ovc lavc -ofps 23.976
-vf pullup
is designed to inverse-telecine
telecined material while leaving progressive data alone. Pullup
doesn't really work well with the current
MEncoder, though, and is really intended
for use with MEncoder G2 (whenever it's
ready). It works fine without -ofps
, but
-ofps
is needed to prevent choppy output. With
-ofps
, it sometimes fails. The problems arise from
MEncoder's behavior of dropping frames to
maintain synchronization between the audio and video: it drops
frames before sending them through the filter chain, rather than
after. As a result, pullup
is sometimes deprived
of the data it needs.
If MEncoder drops too many frames in a
row, it starves pullup
's buffers and causes it to
crash.
Even if MEncoder only drops one frame,
pullup
still doesn't get to see it, and will end
up operating on an incorrect sequence of frames. Even though this
doesn't cause a crash, pullup
won't be able to
make correct decisions on how to reassemble progressive frames, and
will either match fields together incorrectly or drop several fields
to compensate.
I haven't used -vf filmdint
myself, but here's what
D Richard Felker III has to say:
It's OK, but IMO it tries to deinterlace rather than doing inverse telecine too often (much like settop DVD players & progressive TVs) which gives ugly flickering and other artefacts. If you're going to use it, you at least need to spend some time tuning the options and watching the output first to make sure it's not messing up.
There are two options for dealing with this category, each of which is a compromise. You should decide based on the duration/location of each type.
Treat it as progressive. The interlaced parts will look interlaced, and some of the interlaced fields will have to be dropped, resulting in a bit of uneven jumpiness. You can use a postprocessing filter if you want to, but it may slightly degrade the progressive parts.
This option should definitely not be used if you want to eventually
display the video on an interlaced device (with a TV card, for
example). If you have interlaced frames in a 23.976 frames per
second video, they will be telecined along with the progressive
frames. Half of the interlaced "frames" will be displayed for three
fields' duration (3/59.94 seconds), resulting in a flicking
"jump back in time" effect that looks quite bad. If you
even attempt this, you must use a
deinterlacing filter like lb
or
l5
.
It may also be a bad idea for progressive display, too. It will drop pairs of consecutive interlaced fields, resulting in a discontinuity that can be more visible than with the second method, which shows some progressive frames twice. 29.97 frames per second interlaced video is already a bit choppy because it really should be shown at 59.94 fields per second, so the duplicate frames don't stand out as much.
Either way, it's best to consider your content and how you intend to display it. If your video is 90% progressive and you never intend to show it on a TV, you should favor a progressive approach. If it's only half progressive, you probably want to encode it as if it's all interlaced.
Treat it as interlaced. Some frames of the progressive parts will need to be duplicated, resulting in uneven jumpiness. Again, deinterlacing filters may slightly degrade the progressive parts.
About cropping:
Video data on DVDs are stored in a format called YUV 4:2:0. In YUV
video, luma ("brightness") and chroma ("color")
are stored separately. Because the human eye is somewhat less
sensitive to color than it is to brightness, in a YUV 4:2:0 picture
there is only one chroma pixel for every four luma pixels. In a
progressive picture, each square of four luma pixels (two on each
side) has one common chroma pixel. You must crop progressive YUV
4:2:0 to even resolutions, and use even offsets. For example,
crop=716:380:2:26
is OK but
crop=716:380:3:26
is not.
When you are dealing with interlaced YUV 4:2:0, the situation is a bit more complicated. Instead of every four luma pixels in the frame sharing a chroma pixel, every four luma pixels in each field share a chroma pixel. When fields are interlaced to form a frame, each scanline is one pixel high. Now, instead of all four luma pixels being in a square, there are two pixels side-by-side, and the other two pixels are side-by-side two scanlines down. The two luma pixels in the intermediate scanline are from the other field, and so share a different chroma pixel with two luma pixels two scanlines away. All this confusion makes it necessary to have vertical crop dimensions and offsets be multiples of four. Horizontal can stay even.
For telecined video, I recommend that cropping take place after inverse telecining. Once the video is progressive you only need to crop by even numbers. If you really want to gain the slight speedup that cropping first may offer, you must crop vertically by multiples of four or else the inverse-telecine filter won't have proper data.
For interlaced (not telecined) video, you must always crop
vertically by multiples of four unless you use -vf
field
before cropping.
About encoding parameters and quality:
Just because I recommend mbd=2
here doesn't mean it
shouldn't be used elsewhere. Along with trell
,
mbd=2
is one of the two
libavcodec options that
increases quality the most, and you should always use at least those
two unless the drop in encoding speed is prohibitive (e.g. realtime
encoding). There are many other options to
libavcodec that increase
encoding quality (and decrease encoding speed) but that is beyond
the scope of this document.