Report on field synchronisation problems that can cause interlacing artefacts when encoding telecine material from video tape.
What has happened?
We know that for PAL telecine, each film frame is scanned twice to create the two video fields for each video frame. In the first example, where the video frames have no artefacts, the telecine machine advances the film frame during the transition from an F2 field to an F1 field, as shown below:
When the video is encoded to MPEG-2, the video frame is reconstructed using an F1 field followed by an F2 field and the resulting frame exactly matches the original film frame.
In the second example the telecine machine advances the film during the transition from an F1 field to an F2 field. This is shown below:
This time, when the MPEG-2 encoder creates a frame from an F1 field followed by an F2 field, the two fields have been scanned from different film frames. As there are normally spatial differences between the frames, the resulting MPEG video will display comb artefacts.
However, when there is a scene change, you can easily spot that the two fields come from different film frames. This shows that the artefacts are not being caused by the normal interlacing effects you get from video camera originated material.
Why does this happen?
Early PAL telecine machines paid no attention to film frame/video field order. The relationship between film frames and video fields was random until the 1980’s. Vidicon and Plumbicon telecine machines used modified cinema projectors free running into the TV system. 2” VTRs used an 8 field edit sequence to maintain PAL chrominance information, but otherwise the film frame/video field relationship was overlooked or ignored.
Cintel twin-lens and jumpscan telecines ignored field sequence as well, and engineers would devise their own solutions to the problem. Eventually Cintel came up with a device that allowed the engineer to select the relationship between the film frame and video fields.
With NTSC telecine machines the problem of transferring 24 frames per second film to 29.97 frames per second NTSC video has always been far more complex. In order for the transfer to work alternate film frames are scanned twice and then three times in a repeating sequence. The recommended relationship between the film frames and the video fields is described in SMPTE RP 197-2003 and earlier revisions. This technique is commonly referred to as 3:2 pulldown. It has always been recognised that if this sequence is encoded to MPEG video, the duplicated fields will reduce the efficiency of the encoding. To overcome this issue, most MPEG-2 encoders are able to detect and remove the 3:2 pulldown and so reinstate the original film frames.
When asked, most video professionals characterise this problem with PAL telecine as a field order or a field dominance problem and the suggested solutions are to change the field order or to change the field dominance. Unfortunately, the suggested solutions will not work.
Changing field order
Fields always follow an F1 F2 F1 F2 … sequence for both PAL and NTSC so we aren’t talking about the order in which fields are arranged in the video sequence. Field order describes the order in which the F1 and F2 fields are displayed on screen. For PAL video, the F1 field contains the top visible line. This is often called ‘top field first’ order as the as the first field in an F1F2 sequence will contain the top line. For NTSC video, the order is reversed – the F2 field contains the top visible line, so the F1 field will contain the bottom visible line. For obvious reasons this is often called ‘bottom field first’ order.
Some digital systems – early Avid systems for example – always used bottom field first, regardless of the video standard. This was fine for NTSC, but could cause problems for PAL video. No doubt the system was designed by US software engineers, with NTSC in mind!
If we reverse the field order it means that we will display the bottom field as a top field and the top field as a bottom field. Applying this to our PAL telecine material will only make things worse as the fields will not be displayed in the correct spatial relationship to the film frame.
Change field dominance
Field dominance has been described earlier as it relates to video editing. This is a well known issue with video material and most engineers will be able to detect and correct errors where field dominance has been mixed or set incorrectly at the editing stage.
Our problem could be described as a field dominance problem, but unfortunately it is a different issue to the problem that most video engineers describe as field dominance.
If a video tape contains material edited in with mixed field dominance like so:
We have ended up with an odd field from the B clip remaining in the video which may cause a flash. The odd F1 field at the end of the D section won’t cause any visible problem when the video is played back on a TV or video monitor. It would be simple to fix the problem, either by removing the B field F1 and the following D field F2. This would reduce the length of the clip by one frame which is unlikely to be a problem and it has successfully converted the video to F1 dominant video.
Have another look at the way our problem video has been telecine’d:
Each film frame has been transferred as an F2F1 pair of fields. It doesn’t matter where you edit the video, or which field dominance the edit VTR is set to, every film frame will always be an F2F1 field pair. As the PAL standards and recommendations insist that a PAL frame should be an F1 field followed by and F2 field our MPEG encoder is always going to create video frames which mix fields from different film frames.
Field dominance is a good term to describe the issue we have in the sense that we want the telecine machine to produce video that has F1 dominance with respect to the film frames. Unfortunately the term is used in a different sense to describe field dominance with respect to video edits on tape or through switchers, etc. We will use the term ‘telecine field dominance’.
Reversing the telecine field dominance
From the diagram above, it is easy to see how the faulty video could be repaired. Our video stream starts with an F1 field, but the following F2 field is from the second film frame.
Interlacing these two fields creates comb artefacts.
However, the third video field is also from the second film frame. If we interlace the second and third fields, the resulting frame is correct. The second and third fields are spatially different as they are scanned from alternating lines on the film frame. However, these two fields are temporally identical as the film frame doesn’t change between the two scans.
If we swap the second and third fields around, they will be in F1F2 order, which is what the PAL standard expects for a video frame. Similarly with the fourth and fifth frames and all subsequent pairs of frames:
After swapping the second and third and subsequent pairs of fields around our video sequence starts with two F1 fields. This breaks the video standards, so we just discard the first field and effectively lose the first film frame.
There is no readily available hardware that can make this correction. We have used a Windows XP Pro PC with an SDI video capture card, custom software and an SDI video output card to prove that the concept works. We have not yet addressed audio synchronisation although this should be relatively simple.
Some MPEG-2 encoder cards will disguise this problem because of their in-built de-interlacing filters. Some encoders will only encode one of the fields and then interpolate it to create the remaining field. These encoders also disguise the problem. A de-interlacing filter should only be used with video camera originated material. Hopefully, this document will have explained why de-interlacing is not required for PAL telecine material. Discarding a field, means that
half of the video resolution will be lost. Good quality PAL telecine converts perfectly into high quality MPEG files suitable for creating progressive DVDs.
The following two images illustrate this technique applied to faulty video. We have taken the three faulty frames shown earlier and separated them into six consecutive fields. The first and last fields were discarded and the remaining four fields were interlaced to produce two good video frames.
The scene change from our original film has now been reinstated on the telecine’d video. The only evidence of the manipulation is where the frame time codes have been mixed together.
Using relatively simple algorithms it should be possible to detect whether a particular telecine sequence needs to have the reverse field dominance correction applied.
Image quality problems with material being digitised from tape may have its roots in the way the original film material was transferred to video. Early telecine machines had no way for the operator to set the relationship between film frames and video fields. Consequently, the telecine’d material could either start each frame on an F1 field or on an F2 field entirely at random. More recent telecine machines may have a facility to select the relationship between frames and fields, but it must be set correctly.
Because of the confusion between field dominance issues related to editing and field dominance relating to PAL telecine, a telecine engineer may have set the relationship between frames and fields according to their own preference. It would have been better if the telecine machine did not allow for adjustment and always advanced the film on the transition between an F2 and an F1 field.
This problem is going to become more and more prevalent as film archives are transferred to the digital domain. If the telecine transfer is done at the same time as the digital transfer then the telecine process can be configured to produce correct results consistently. For archives that were telecine’d some time in the past, up to 50% of the material may need to be corrected before being digitised.
There are no easy fixes to this problem – perhaps because it is not an issue while the telecine material remains in the interlaced video domain. None of the well known video equipment manufacturers produce a black box to apply the fix discussed earlier – reversing the order of pairs of fields is unusual after all.
Without a computer-based or black box device that can reverse fields there are only two ways to proceed. The telecine material can be encoded using a single field – this will remove the comb artefacts at the expense of losing half the picture information. Alternatively, both fields can be used to create an MPEG file which can be used as is, or it could be processed further to correct the problem. If the MPEG file is processed further, there will be a generational loss and increased compression artefacts in the final MPEG file.