Telecine 3:2 Pulldown

http://www.dvdfile.com/news/special_report/production_a_z/3_2_pulldown.htm

by Dan Ramer

As we've highlighted hardware technologies at DVDFILE, we've touched upon progressive scan DVD players and line doublers that convert interlaced video to progressive video. Some home theater components that produce progressive scan video have the commendable ability to correct for most of the unavoidable visual distortions caused by the 3:2 pulldown process. Or tech editor was so impressed with the Toshiba progressive scan DVD player he reviewed that it became his reference unit. So let's take a look at the 3:2 pulldown process, discuss why it's necessary, describe the artifacts, and consider how we might get rid of its visual distortions.

You wouldn't have to read this article if film and video weren't so different. It's due to those differences that the 3:2 pulldown process becomes necessary. So to begin, we should examine the nature of video and film.

NTSC (National Television Standards Committee) Video is composed of 525 horizontal scan lines. In our wonderful world of DVD, 480 of those scan lines are available to contain picture information. NTSC video is interlaced. In other words, even though the video is shown at 29.97 frames (pictures) per second, each video frame contains two video fields. One field is composed of all the odd horizontal scan lines; the other contains all the even horizontal scan lines. So despite the reality that NTSC video displays 29.97 frames or pictures each second, it's actually created as 59.94 fields per second. Consequently, any four sequential video frames (A, B, C, D) are drawn on the video display as A1, A2, B1, B2, C1, C2, D1, D2, where the 1 or 2 represents the field number within the frame. This is what the vast majority of video displays - including, most likely, yours - expects from any video signal source.

Conventional 35 mm and 70 mm film is shot at 24 frames per second. On the motion picture screen, visible flicker is minimized by projecting the film at 48 frames per second. To maintain proper speed onscreen, the projector repeats each frame. So any four film frames would be projected as A, A, B, B, C, C, D, D. Since the frame rates of film and NTSC video are quite different (24 film frames per second as opposed to 29.97 video frames per second), when we transfer film to video or try to display film from video, we have a bit of a problem.

Simply transferring each film frame onto each video frame would result in a film running about 24.9% faster than intended; 29.97 film frames would be shown during each second rather than the correct 24. The clever solution to this problem is to repeat film frames periodically in a very straightforward but mathematically significant way; the resultant redundancy prevents the apparent speedup of the film when shown at the conventional video frame rate. This is how it's done.

The telecine machine used to transfer film to video for composite D2 masters (which may be used for VHS cassettes, laserdiscs, and broadcast) projects film onto a video imager at 59.94 frames per second (identical to and synchronized with the video field rate) and repeats film frames in a recurring 3:2 pattern. In other words, the film frame sequence is A, A, A, B, B, C, C, C, D, D, and so on:

The first film frame, A, is repeated three times and is recorded as field 1 and field 2 of the first video frame, and field 1 of the second video frame. The second film frame, B, is repeated twice and is recorded as field 2 of the second video frame and field 1 of the third video frame. The third film frame, C, is repeated three times and is recorded as field 2 of the third video frame and fields 1 and 2 of the fourth video frame. The fourth film frame, D, is repeated twice and is recorded as field 1 and field 2 of the fifth video frame. See the pattern?Repeat this sequence six times and 24 frames of film become 30 frames of video.

So the basis of this technique is to restore proper timing by generating redundant image information from four film frames within every five NTSC video frames. But wouldn't it be silly to waste 20% of the storage space on every DVD with duplicate picture data? Fortunately the MPEG-2 standard nicely avoids this inefficiency. When a film source is encoded for presentation on DVD, it is stored at 24 frames per second; each video frame contains all the picture information from each film frame. There is no redundancy or duplication. Such a transfer is written to DVD as 720-pixel wide by 480-pixel high interlaced frames (where each frame contains two 720 by 240 fields), and there are only 24 frames for each second of film. This is known as 480i24. On each DVD encoded from a film source, a flag is inserted within the MPEG-2 data stream that instructs the player to repeat certain fields to reconstruct the 29.97 frame per second interlaced video. The player obliges by performing the 3:2 pulldown in real-time, continually creating interlaced frame sequences just like the one shown in the above figure, "The Telecine 3:2 Pulldown Process for NTSC Video." This capability enables the player to produce video compatible with conventional displays that were designed based on the NTSC video standard. (As we shall see later, progressive scan DVD players take a different approach.)

While the 3:2 pulldown process restores the proper speed of the film on video, it generates some unpleasant problems. Two sequential video frames within every five video frame sequence contain images from different film frames. If there is movement of the images on film, 40% of the video frames will contain visually distorted information. Let me steal a figure from my Anamorphic Widescreen piece to demonstrate.

The video frame on the left is fine. The circle on film was quite still and so the odd and even scan lines paint a stable video picture. Now let's pan left on film, causing an apparent motion of the circle to the right. Notice that within the video frame on the right - one of the two frames in the five video frame sequence that contains fields from two different film frames - the circle is in one position based on the odd scan lines and in another position based on the even scan lines. What a mess. Repeat this process 40% of the time and the eye sees a loss of focus, a smearing of detail, for any moving object. For those frames that contain a quick cut from one scene to another, the image may become even odder:

Here, the film editor has cut from our image of the black circle to images of a green rectangle to the right and part of a blue cone to the left. For a video frame that captures images from these two different film frames, one before and one after the scene change, all three objects appear on the video display for the duration of the video frame. For that brief snatch of time (33.37 msec), our vertical resolution has been cut in half for that film frame.

These are the two spatial artifacts caused by interlacing; I'll touch upon a temporal artifact soon. Now, let's see if we can minimize these spatial flaws as we watch our DVDs.

Let's start with the solution that's been around the longest, reverse the 3:2 pulldown as the video is converted within a line doubler from interlaced video to progressive video:

Each progressive video frame is reconstructed by weaving together the odd and even fields from images that were derived from the same film frame. The video frames are then shown at double the conventional NTSC video frame rate in a 3:2 repeating pattern. This effectively doubles the number of horizontal scan lines during each second, hence the name of the instrument that performs the work: a line doubler.(It isn't clear whether C2 is derived from interlaced frame 3 or interlaced frame 4; I arbitrarily showed frame 4.)The technically astute might notice that this scheme seems to require going forward in time to reconstruct frame B. Actually, frame buffers and double buffering techniques are used to overcome this problem. This implies that there is a slight delay through such complex video circuitry, but experience has shown that synchronization with non-delayed audio is not an issue.

While lacking strict temporal consistency - every other film image is on the screen for one and a half times longer than the previous film image causing an apparent subtle jerking or juddering during smooth scans - the spatial distortion of interlacing fields from two different film frames is gone. Since the reverse 3:2 pulldown is somewhat complex, requiring some field analysis on the fly to get it right, it's not available in all line doublers. You'll find this feature in the surprisingly inexpensive DVDO line doubler and such high-end video processors as those from Faroudja. Variations on this theme may be found in some quadruplers, interpolators, and scalars, but that's a topic for some other time.

As I mentioned earlier, film is stored on DVD as 480i at the equivalent of 24 frames per second. When a conventional player recognizes the appropriate MPEG-2 frame repeat flag, it performs the 3:2 pulldown in real-time, but progressive scan players can react to this flag in a different way. Such a player can create progressive video in real-time .It reconstructs each video frame by weaving together its odd and even fields, then repeats the video frames in a recurring 3:2 pattern. The resulting video signal will contain the same frame sequence and the same horizontal and vertical scan rates as are produced by the line doubler. This is a simpler process than is required in a line doubler since the player does not have to examine the fields to determine how to perform the weaving; no DVD derived from film contains a video frame with images from two film frames.

One potential advantage of performing this process within the DVD player is that it's done entirely in the digital domain, so no signal degradation occurs. An external line doubler accepts a DVD's video signal in analog form, such as component or S-video. The line doubler must digitize the video to bring it into its digital processing circuitry. The line-doubled digital video is then transformed to analog once again for compatibility with the video display. With no less than an analog buffer, an anti-aliasing filter, a sample-and-hold, an analog-to-digital converter, a digital-to-analog converter, another anti-aliasing filter, and another analog buffer involved in the conversions from analog to digital to analog, there's quit a bit of circuitry that can get in the way of a pristine signal. Only the most expensive video processors, costing thousands of dollars, will perform these tasks without visibly degrading the video.

Please note that for a video display to properly present such progressive video or line-doubled signal, it must be capable of dealing with about 31,500 scan lines per second - twice the normal rate. Interestingly, the vertical sync rate remains the same as conventional NTSC video, 59.94 Hz.

Because many computer displays are capable of broader ranges of horizontal and vertical scan rates, it is possible to create temporally symmetrical progressive video that runs at two or three times the film's frame rate: 48 or, more commonly, 72 frames per second. To maintain proper timing, each frame must be repeated two or three times, respectively, so the sequence becomes A, A, B, B, C, C, D, D or A, A, A, B, B, B, C, C, C, D, D, D. Each film image is shown on the video display for precisely the same amount of time, creating our temporal symmetry. So not only have we eliminated the spatial distortions, juddering during smooth pans is now gone as well.

48 frames per second require a horizontal scan rate of 25,200 Hz and a vertical sync rate that extends down to 48 Hz. 72 frames per second require a horizontal scan rate of 37,800 Hz and a vertical sync rate that extends up to 72 Hz. Interestingly, many front projectors are capable of these rates. I've received e-mail from home theater enthusiasts who prefer to use their computers as DVD players to take advantage of this flavor of progressive scan on such projectors. I suspect that as more capable video displays become readily available, we may see standalone progressive DVD players that offer the 48 or 72 frame per second playback option.

It's quite remarkable how much the image quality can be improved by eliminating 3:2 pulldown artifacts with an appropriate reverse process while converting to progressive video. Throw in a good anamorphic transfer with no edge enhancement and the presentation is surprisingly film-like. Interested? You can expect progressive scan DVD players to be introduced by several manufacturers this year. Or you might want to investigate a line doubler, perhaps the affordable DVDO. And as HDTV-ready display prices come down, more and more of you will be able to enjoy the best home theater currently has to offer.

http://ubuntuforums.org/showthread.php?t=193754&page=3

Here are few of my command line options for *DVD*s:

Explanation:

This is NTSC encoded file with 3:2 pulldown. Means that file is encoded at 23.976 fps (24000/1001), while dvd player is tricked to think it is 29.97 (that's what 'telecine' option does in mpegopts). This way we have much smaller file with same quality. For PAL encodings use '-ofps 25' and exclude 'telecine' in mpegopts.

In lavcopts we have some settings set to achieve very high quality encodings. You can change abitrate to ie 224 if you want higher quality sound. Option vqscale sets constant quantizer (4 in this case), lower the quant, better the quality, but also bigger file. Quantizer 1 is not recomended (compatibility issues). If you want fixed size file, you should use 2-pass encoding.

Video filters ('-vf' option) sets some filters. Include 'al' (autolevels) only if your video source has washed-up look. 'scale' settings are set to give 2px border around video (giving smaller size encoding, those 2 pixels are not visible on most tvs).

Note that those settings are tweaked for widescreen material, you should change few things (like autoaspect option in lavcopts), but I did not bother to check. Also '-of mpeg' is still in experimental stage, so output file might need to be remultiplexed (this issue might be solved in mencoder pre8 that's been released recently). Read man for more info.

Recently I've discovered that transcode with mpeg2enc gives even better results than mencoder, only it can be very slow sometimes. Source is also ntsc pulled down to 29.97. Some cool options are in '-J'. First - filters used are mplayer ones, same as mecoder; do not use 'al' unless for washed-up videos. 'normalize' will normalize your sound. 'text' is the easiest way to add text to your video (range is frame start/end).

Again change '-b' option to 224 if you want higher quality sound.

Once encoding is finished, you need to multiplex video/audio files:

Transcode will ensure that you have very compliant dvd video with superb quality.

Last edited by encho : July 5th, 2006