Coeus wrote: ↑
Sun Jul 01, 2018 12:03 pm
how far wide of the mark would I bet in suggesting that a de-interlacing step would be to know whether the field concerned is odd or even, interpolate the missing lines from the ones present so the video is now 50fps at the full vertical resolution and then process further as desired?
It depends on the format in which the interlaced video arrives. If it has been assembled as if it was 25p, you don't need to know which field was originally 'odd' and which 'even' because the spatial
relationship can be inferred from the order of lines in the combined frame; in that case what you need to ascertain is the temporal
order of the two fields (i.e. which originally came first).
However if the interlaced video arrives as separate fields
(less common) you are in a different situation. Now the temporal order is known, but the spatial relationship between the fields isn't. One way of establishing that spatial relationship is knowing which was originally 'odd' and which 'even' - although complications such as 'half-lines' and precisely the length of vertical blanking may result in that being unreliable.
Either way, in order to create the required 50p output (and you're right in saying it should be 50p, not the 25p in which the CLPA material has been made available) you need to know both the temporal and spatial relationships between the incoming lines you are presented with.
am wondering if the issue is that one doesn't need any kind of technical background to use modern video processing tools on a PC or Mac.
I suspect that is indeed part of the problem. I know that the CLPA people were influenced by this GitHub paper
which asserts that Martin Weston's de-interlacing filter - which is the one preferred by the BBC - "didn't work properly". Clearly this must have been as a result of some misuse or misunderstanding, since the filter is known to be excellent, but it caused it to be rejected by CLPA.
Many years ago I wrote an Aspect Ratio Converter (ARC) program in BBC BASIC, which uses the Weston filter (when performing any vertical scaling operation you must first de-interlace, then perform the scaling, then re-interlace if necessary). That program makes no assumptions about the temporal relationship between the input fields but instead determines it automatically using a correlation process (only possible if there is enough motion present in the source material of course).