Pull-up (video conversion)
The use of language between the terms pull-up [ pʊlˈʌp ] and pull-down has not been clearly clarified. The conversion of film material with 24 or 25 full frames / second into 50 or 60 fields / second suitable for classic television transmission is sometimes referred to as 3: 2 pull-down or 2: 3 pull-up .
Sometimes the source material is also accelerated or slowed down a little in order to reach the target number of images: The conversion of 24 Hz full image to 25 Hz full image (50 Hz half image) by approx. 4% faster playback is called PAL acceleration , the ( rare) reverse process called "PAL Speeddown".
In addition to the format conversion from full-image formats to field formats (non-multiple frequency), there are also processes that reassemble (pseudo) fields that belong together to form full images; in some cases, fields that are simply transmitted twice are discarded (or used to reduce noise). These processes are necessary before any post-processing of images on the display side.
Compounding is only possible without loss if the source material was full images that were split directly into half images (and possibly repeated). The veneering technique , which is also frequently used , in which images are mixed, or if the field material was filmed directly as fields, complicated processes must be used.
See v. a. Deinterlacing , Inverse Telecine .
Calculation of intermediate images for LCDs
Since all currently available panels refresh the display internally with 60 images / second, a conversion of the image to 60 full images per second is necessary for a clean moving display without double contours and jerking. This conversion is required in all modes that do not transmit 60 frames per second, i.e. H. for practically all sources. This conversion is done by generating intermediate images, which are calculated by recognizing movements in the image. The algorithms used for this and the necessary processing services are almost identical for images with and without interlacing, but when interlacing, the even and odd lines of the internally calculated image are updated at different times with actually sent images.
Conversion from 25p to 50p:
Read frame n + 0: A (n + 0, x, y) ────────────── ▶ output of A (n + 0, x, y)
│
│ ↑
│
Search of motion vectors v (n + 0, x, y) ───── ▶ Output of ½ A (n + 0, x, y) ° ½ v (n + 0, x, y) + ½ A (n + 1, x, y) ° -½ v (n + 0, x, y)
│
│ ↓
│
Read frame n + 1: A (n + 1, x, y) ────────────── ▶ output of A (n + 1, x, y)
│
│ ↑
│
Search for motion vectors v (n + 1, x, y) ───── ▶ Output ½ A (n + 1, x, y) ° ½ v (n + 1, x, y) + ½ A (n + 2, x, y) ° -½ v (n + 1, x, y)
│
│ ↓
│
Read frame n + 2: A (n + 2, x, y) ────────────── ▶ output of A (n + 2, x, y)
│
│ ↑
│
Note: The "°" symbol stands for the shift operator.
Conversion from 24p to 60p:
Read frame n + 0: A (n + 0, x, y) ────────────── ▶ output of A (n + 0, x, y)
│
│ ↑
│
Search of motion vectors v (n + 0, x, y) ──┬── ▶ Output of 0.6 · A (n + 0, x, y) ° 0.4 v (n + 0, x, y) + 0.4 A (n + 1, x, y) ° -0.6 v (n + 0, x, y)
│ │
│ ↓ └── ▶ Output of 0.2 A (n + 0, x, y) ° 0.8 v (n + 0, x, y) + 0.8 A (n + 1, x, y ) ° -0.2 v (n + 0, x, y)
│
Read full screen n + 1: A (n + 1, x, y)
│ ┌── ▶ Output of 0.8 A (n + 0, x, y) ° 0.2 v (n + 1, x, y) + 0.2 A (n + 2, x, y) ° -0.8 v (n + 1, x, y)
│ ↑ │
│ │
Search of motion vectors v (n + 1, x, y) ──┴── ▶ Output of 0.4 · A (n + 0, x, y) ° 0.6 v (n + 1, x, y) + 0.6 A (n + 2, x, y) ° -0.4 v (n + 1, x, y)
│
│ ↓
│
Read frame n + 2: A (n + 2, x, y) ──────────────> output of A (n + 2, x, y)
│
│ ↑
│
Note: Here every second image is not output at all (unprocessed).
.png)
If the source is interlaced, the functions "Read full screen" and "Find motion vectors" are modified. Reading a field is done by shifting the previous field and replacing the lines that were sent in this field. The search for the motion vectors is carried out partly by pattern search in the previous field and partly by pattern search in the previous field.
See also: Moving images , film scanners