Inverse Telecine
Inverse telecine reconstructs progressive film pictures from a telecined television cadence.
A film transfer can store five television field pairs for every four original film pictures. Inverse telecine (IVTC) recovers those original pictures and removes the redundant presentation created by pulldown.
The recovery has two distinct operations: field matching reconstructs complete pictures; decimation removes duplicate presentations. Keeping them separate makes the process understandable.
Start with the stored cycle
Use the two-field-phase sequence from the telecine example:
| Pair | Top field | Bottom field | Woven result |
|---|---|---|---|
| 0 | A, field 0 | A, field 1 | A |
| 1 | B, field 2 | B, field 3 | B |
| 2 | B, field 4 | C, field 5 | Mixed B/C |
| 3 | C, field 6 | D, field 7 | Mixed C/D |
| 4 | D, field 8 | D, field 9 | D |
The mixed pairs have enough data in their neighborhood to recover film pictures, but their own top and bottom fields do not belong together.
Match rows from the same film picture
A suitable matching phase can construct:
| Reconstructed picture | Fields selected | Why they match |
|---|---|---|
| A | Top 0 + bottom 1 | Both came from A |
| B | Top 2 + bottom 3 | Both came from B |
| C | Top 6 + bottom 5 | Both came from C, across a pair boundary |
| D | Top 8 + bottom 7 | Both came from D, across a pair boundary |
| D again | Top 8 + bottom 9 | Another presentation of D |
The candidate result is A, B, C, D, D. Another matching phase can put the duplicate elsewhere. The desired result is the same four source pictures.
This is reconstruction from existing rows. Ordinary deinterlacing would instead estimate rows missing at a selected field time. For this clean film sequence, matching can retain the source picture's line detail because both row sets actually exist.
In real compressed transfers, repeated fields can differ slightly after encoding or processing. Duplicate detection may therefore use similarity measurements rather than expecting identical file bytes or perfect pixel checksums.
Remove redundancy and rebuild the timeline
Drop the redundant D, leaving A, B, C, D.
At 30000/1001 frame pairs/s, the five-pair cycle lasts:
5 × 1001/30000 = 0.166833… seconds.
At 24000/1001 progressive pictures/s, four pictures last:
4 × 1001/24000 = 0.166833… seconds.
The picture count falls, but the programme's duration stays the same. Each recovered picture is now assigned its proper film duration instead of retaining the five-slot television schedule. Audio can remain on the original programme timeline.
The operation is therefore more than deleting every fifth frame. Matching fixes the picture content; appropriate timestamps fix its presentation.
What an edit changes
An edit can begin a new cadence phase. A field matcher needs to examine neighboring candidates rather than assuming one phase for the whole file. A mismatched pair near a cut can combine two entirely different scenes, so the transition itself needs attention.
FFmpeg's fieldmatch and decimate illustrate these separate stages. Field matching reconstructs pictures but does not drop duplicates. A deinterlacing fallback can handle pictures that remain combed before a later allocation stage.
Mixed material introduces another decision. A programme may have telecined film plus a native video insert or a moving graphic. In the native section, successive fields contain new motion, not repeated film rows. Forcing a film-rate removal pattern there throws away temporal information.
A recovery workflow can treat sections separately, retain variable picture timing, or choose an output schedule capable of representing both. What matters is following the capture relationships and the timeline of each section, rather than requiring every part of the programme to fit one cadence.
When reviewing a restored Emby source, check clean moving frames across edits and the programme's duration. Those observations show whether both matching and presentation have been recovered.