Problems of Automatic Inverse Telecine of NTSC Video

As I hinted earlier, automatic inverse telecine of an AVI file isn't perfect.  Sometimes this is the fault of the inverse telecine program, while at other times the problem lies in the video itself.  The result of incorrect inverse telecining is a frame of video that contains interlacing.  The interlacing appears because the wrong fields were joined while trying to reconstruct the original film frame.  

What Happens When a Program Incorrectly Inverse Telecines a Full Resolution Video?

When a program doesn't inverse telecine a full resolution video (XXX by 480) correctly, one or two fields that aren't duplicates are inappropriately discarded.  Because the duplicate fields are left intact, interlacing artifacts will be present in your video.  Here are some diagrams that explain why this occurs:

Source Video (The underlined fields are duplicates.)

The correct way to inverse telecine the source frames would be to remove the top field of TV Frame #2 (1T) and the bottom field of TV Frame #2 (3B).  However, sometimes the inverse teleciner messes up and doesn't choose those fields.  For example, it might choose top field of TV Frame #3 (2T) and the bottom field of TV Frame #4 (3B).  This leads to the following results:

In the above picture, the top field of TV Frame #3 (2T) and the bottom field of TV Frame #4 (3B) were incorrectly discarded.  This left the actual duplicated field, 1T and 3B, completely intact.  Because of this, the video will contain interlacing artifacts in frame B when viewed on a computer.  Additionally, half of the film frame 2 is lost, which represents a loss of detail.

What Happens When a Program Incorrectly Inverse Telecines a Low Resolution Video?

When a program doesn't inverse telecine a low resolution video (XXX by 240) correctly, two virtual fields that don't belong together are inappropriately combined.  This actually creates interlacing artifacts in a video where none was present!  Here are some diagrams of what I mean:

Source video (the underlined field is a duplicate)

Because the video was captured at low resolution, it only contains one field.  However, like I mentioned earlier, the source video is seen by the inverse telecine program as being made up of two virtual fields.  All the odd horizontal lines in the video are treated like a top field and all the even horizontal lines are treated like a bottom field.  Therefore, the video looks like this:

Source video as seen by inverse telecine program

The correct way to inverse telecine the source frames would be to remove the top "field" of frame #1 and the bottom "field" of frame #2.  However, sometimes the inverse teleciner messes up and doesn't choose those "fields".  For example, it might choose the top "field" of frame #2 and the bottom "field" of frame #3.  This leads to the following results:

In the above picture, the top "field" of source frame #2 and the bottom "field" of source frame #3 were incorrectly discarded.  When the remaining "fields" were combined to produce the new frame B, the "fields" did not match.  Because of this, the video will contain interlacing artifacts when viewed on a computer.  Additionally, half of the film frame 2 is lost, which represents a loss of detail.

Why Do Programs Mess Up When Doing an Automatic Inverse Telecine?

There are actually a number of factors that can cause a program to incorrectly inverse telecine a video.  I will describe some of the more common ones below.

1)  Inaccuracy of programs that inverse telecine.

An automatic inverse telecining program must determine which fields are duplicated and consequently remove them.  However, this task is not as easy as it sounds.  It is very challenging to come up with a computer algorithm that accurately finds the duplicated fields.  For example, when there is little motion in a scene, it is difficult to determine which frames are the duplicates.  The mere presence of video noise can make two duplicated frames appear to be unique.

2)  Shifts in the telecining pattern.

Most videos are not recorded at 24 fps and then converted to ~30 frames per second when the video is completed.  Instead, the video is recorded in many different parts, each of which are telecined to ~30 fps separately.  When the parts are put together, the telecining pattern doesn't always line up.  For example, let's say that Part 1 of a video ended on the first interlaced frame in a group of 5 frames.  Now let's say that Part 2 of a video begins on the first progressive (non-interlaced) frame in a group of 5 frames.  If we were to join Part 1 to Part 2, there would only be one interlaced frame in between the progressive frames.  This would lead to a telecining pattern that looks like this:

In the above picture, Part 1 ends at frame #4 and Part 2 begins on frame #5.  If you look carefully, you'll notice that only one field of film frame 4 (4B) is present in the video.  Because of this, it's actually impossible to reconstruct the original film frames.  The best you could hope to do is deinterlace frame #4 to keep only field 4B, but this process is difficult and half of the detail will still be lost.

It's also possible for two parts of a video to be joined together between progressive frames.  For example, the joined video might look like this:

Part 1 ends at frame #2 and Part 2 begins at frame #3.  As you can see, there are 5 progressive frames followed by 2 interlaced frames, rather than 3 progressive frame followed by 2 interlaced frames.  If you're lucky, the inverse telecining program will detect the shift in the telecining pattern and not remove any fields from frames #1 through #5.  However, since the program is expecting interlaced video to be present in frame #4, it might start discarding fields there even though it shouldn't. 

3)  Frame drops.

If you captured video has dropped frames, it can cause the inverse teleciner to mess up.  A dropped frame looks like a temporary shift in the telecining pattern to a program that inverse telecines.  It is handled in the same fashion as shifts in the telecining pattern caused by bad joins.  As a result, reconstructing the original film frames can be difficult or impossible.

4)  Different telecine patterns playing at the same time.

Sometimes, more than one video is being played on a television at once.  For example, during remote satellite interviews they often show the interviewer talking at location A while the interviewee is talking at location B.  If these two videos have different telecining patterns, it is impossible to correctly inverse telecine the program.  This is because the inverse teleciner assumes that the whole frame has a certain telecining pattern.  Having more than one video on screen produces different telecining patterns in different parts of the same frame.