Interlacing, Combing, and Old Footage

Interlaced video, the format behind MiniDV tape, AVCHD camcorder MTS clips, and every DVD, draws a frame from two separate instants: one set of lines from a moment now, the rest from a moment a fraction of a second later. On a screen expecting one instant per frame, anything that moved in between shows up as teeth. We built a comb artifact the way a real camera would and measured what it costs a video, in quality and in bytes, to leave it alone versus to fix it.

Why does my video have horizontal lines?

Because it's interlaced, and something in the pipeline drew both of its fields as one static frame instead of alternating them the way a CRT or a deinterlacer would.

• Teeth on anything moving? That's combing. Deinterlacing removes it; changing the container or codec alone doesn't.
• Converting DV tape captures or AVCHD camcorder clips? Convert DV to MP4 or convert MTS to MP4 for compatibility, and read the deinterlacing section below first if the source shows teeth.
• File plays fine, nothing looks combed? It's probably progressive already, or was deinterlaced before you got it.

Interlacing at a glance

What's the difference between a field and a frame?

A frame is a complete picture. A field is half of one, the odd-numbered scan lines or the even-numbered ones, and interlaced video sends two per frame interval instead of one full picture.

Adobe's own documentation on interlaced video states it plainly: "only half the number of horizontal lines for each frame of video are transmitted at a time," the upper field's odd lines drawn first, the lower field's even lines right after. Broadcast engineers chose this in the analog era so a fixed bandwidth could carry motion at double the rate, at the cost of half the vertical detail in each field.

A television built for the signal drew the two fields as close-together refreshes that persistence of vision blended into one picture. A modern screen has no such refresh cycle; it wants one full frame per interval, so software either weaves the two fields into a single frame, which produces combing on anything that moved, or deinterlaces them first. The same Adobe page lists "combing artifacts when motion occurs between fields" among the format's drawbacks, alongside flicker on thin horizontal detail.

Where do you still run into interlaced footage today?

Mostly MiniDV tape captures, AVCHD camcorder clips, and DVDs, all formats built when interlaced broadcast standards were still how a television displayed anything.

DV, what MiniDV, DVCAM and DVCPRO camcorders wrote to tape from 1995, ties its frame rate to the region's broadcast standard: 29.97 interlaced fields a second for NTSC, 25 for PAL. MTS, the extension AVCHD camcorders write since 2006, was built around the same 1080i structure at launch, though later cameras added progressive 1080p and 720p modes. VOB, inside every commercial DVD since 1996, is interlaced MPEG-2 tied to the same field rate.

None of this means every file you find in these formats is interlaced. AVCHD's progressive modes and some prosumer DV decks recorded 25p or similar without fields at all. The only way to know for certain is to look, which the questions section below covers.

What does a comb artifact actually look like?

Jagged, staircase teeth along any edge that moved between the two fields drawn into that frame, not blur, a literal one- or two-pixel misalignment repeating down the edge.

We don't have a real interlaced camera in the corpus, so we built one the way a real sensor would: FFmpeg's tinterlace filter wove the upper field of one frame of a progressive test clip with the lower field of a later frame, the same operation an interlaced sensor performs across two exposures.

That's the mechanism. What it costs, measured at a realistic one-frame gap rather than the exaggerated one above, is next.

What does deinterlacing actually cost you?

Some quality either way, and a choice between half your original frame rate or roughly double the encoding work, depending which deinterlacing mode you pick.

We wove the same source clip's fields together with a one-frame gap, matching an interlaced camera, deinterlaced the result two ways with FFmpeg's bwdif filter, and measured VMAF against the untouched source, all encoded through the same codec, CRF and preset so the only variable is what happened to the fields:

Frame mode blends each pair of fields into one frame, so the clip stays stuck at half its original frame rate, recovering most of the quality the comb cost, 93.8 against a 96.0 ceiling, but choppier motion. Bob mode builds a full frame from every individual field instead, getting your original 30 fps back at a slightly lower 91.0, since each frame only had real detail in half its lines to start with.

Left alone, combed footage even costs more bytes than fixing it does: a combed frame took 126,481 bytes at matched CRF 18 against a clean frame's 54,249, more than double, because the stair-step pattern is high-frequency detail an encoder can't predict away. Interlacing doesn't just look worse when it's mishandled. It compresses worse too.

What happens when you also need to change the frame rate?

Converting old film footage to video needs more than a frame-rate number typed into a box, because film runs at 24 frames a second and NTSC video needs 29.97.

Apple's own documentation on 3:2 pulldown describes the NTSC pattern directly: "if you alternate recording two fields of one film frame and then three fields of the next, the 24 frames in 1 second of film end up filling the 30 frames in 1 second of video." PAL needs no such trick; 25 fps splits evenly into 50 fields.

This matters because telecined footage is recoverable in a way plain interlaced footage isn't: a telecined frame is a repackaged film frame, so removing the repeated fields, inverse telecine, gets the original 24 fps picture back with no loss. Ordinary interlaced footage has no such frame hiding inside it; the missing half of each field's detail was never captured.

FileFlip's Frame rate control just changes how many frames the output has. We tested what that does to already-combed footage alone: stretching the still-combed clip above from 15 to 30 fps with nothing but a frame rate change scored VMAF 86.8, worse than leaving the comb untouched at 89.8, since every duplicated frame repeats the same tear twice.

What do you lose when converting old interlaced footage?

Nothing about the comb pattern. FileFlip's Advanced options for a video conversion offer Resolution, Bitrate and Frame rate, and none of them run a deinterlacing filter, so a combed frame going in comes out just as combed, whether the operation ends up a remux or a full re-encode; see remuxing vs transcoding for why changing a container or codec never touches what's baked into a frame's pixels.

If your DV or MTS source is real interlaced camera footage and the comb bothers you, deinterlace it in an editor first, Premiere Pro, Final Cut Pro and DaVinci Resolve all ship a deinterlace effect, then convert the result.

Common questions

Is DV or MTS video always interlaced?

No. AVCHD added progressive 1080p and 720p recording after its 2006 launch, and some prosumer DV decks recorded 25p without fields. Interlaced was the norm for consumer MiniDV and early AVCHD, but "DV" and "MTS" name a container and codec, not a guarantee about fields.

How can I tell if my video is interlaced?

Step through it frame by frame and look at something moving. A progressive frame stays sharp; an interlaced frame shows teeth on the moving edge and nowhere else. ffprobe -show_entries stream=field_order also reports the flag directly when a file carries one, though not every file sets it accurately.

Does converting DV or MTS to MP4 fix interlacing on its own?

No. A conversion changes the container and, when needed, the codec; it doesn't touch how the fields inside each frame were drawn. See remuxing vs transcoding for the same fact applied to file size instead: changing the wrapper around a picture never changes the picture.

Is telecine the same thing as interlacing?

No, though telecined footage is interlaced. Interlacing is how a signal transmits two fields per frame interval; telecine is the specific process of fitting 24 fps film into a 29.97 fps interlaced signal with the 3:2 pattern above. A telecined frame reverses back to the original film frame exactly; ordinary interlaced footage cannot.

Convert between them

FileFlip converts DV to MP4 and MTS to MP4 entirely in your browser, on a WebAssembly build of FFmpeg. Nothing uploads, there's no account, and the codec and container change while field order does not. If the source shows combing and you want it gone, deinterlace it before you convert; FileFlip works with whatever pixels the file already has.

For general background on how video codecs work, see how video compression works.

How we measured this

• Source clip: the 10-second, 1920x1080, 30 fps progressive H.264 Big Buck Bunny segment (Blender Foundation, CC BY 3.0) from remuxing vs transcoding, fetched with pnpm blog:corpora.
• Simulated interlacing: FFmpeg's tinterlace=mode=interleave_top, weaving the upper field of one frame with the lower field of the next, unchanged height, half the frame rate, the standard way to build interlaced test material from a progressive source.
• Deinterlacing: FFmpeg's bwdif filter, mode=send_frame (one output frame per input frame) and mode=send_field (one output frame per field).
• Frame rate change with no deinterlacing: ffmpeg -r 30 on the combed clip, the operation behind FileFlip's Frame rate control.
• Encoding: every version through libx264 -crf 18 -preset medium, held constant so the only variable is what happened to the fields; a plain re-encode with no interlacing step is the control, scoring VMAF 96.0.
• Quality: FFmpeg's libvmaf filter, against the source for full-frame-rate outputs, or against a reference built from only the source frames supplying each combed frame's upper field for half-frame-rate outputs.
• Tools: FFmpeg 8.1.2 with libx264 and libvmaf, on macOS.
• Caveat: one 10-second clip is a single data point, and the figure above uses a wider gap between the two woven frames than the one-frame gap the tables use, so the comb reads clearly at this crop.
• Reproducing it: node scripts/benchmark-interlacing.mjs <clip> prints every row of the cost table above.