How to remove grain, how to fix film shot in low light?

[lordsmurf]

This is what I see in the 10 samples you've posted:

• There is mosquito noise from compression, and it appears to be MPEG noise, not H.264 noise. My suspicion is that it was captured to DVD/MPEG first, and then transcoded to H.264 next. What I see there is not normal.
• The deinterlace used was Yadif. Not terrible, but not the best. Yadif gets overrun with stairsteps in linear objects like air vents and house siding. Sometimes, the deinterlace job in the clips actually looks like older adaptive, not newer/better Yadif, so I'm not 100% which was used. It was definitely not QTGMC.
• To process video (deinterlace, H.264, etc) and not cover overscan with a matte, is honestly inexcusable. I understand seeing overscan on raw captures, but never post-capture.
• Those samples all shake too much to really tell if TBC was used. I think it was, as non-TBC conversion is generally much worse, but sometimes I see furniture and I'm not sure if it is moving. A good transport VCR can minimize no-TBC wiggles, but never fully remove/prevent them.
• Tape grain is present, and was not removed. The shaky nature of the film will not react well to temporal NR methods, and non-temporal methods will primary blur. And VHS footage is already soft as is. If you remove too much of the grain, I'm afraid that the poorer deinterlace method would be more obvious.
• The mosquito noise is treatable with some Avisynth filtering, perhaps deen() or mosquitoNR() -- or maybe another filter, as Avisynth is complicated primarily because it has seemingly endless options, and no two people ever use it the same way.

Overall, I've seen worse.

What they did for you is decent, but not what I'd call good.
Better quality is easily possible, but most services are sadly not the ones to give it. That's why it's really important to do research about video, and hardware used, before going with them.
I could do better without really even trying.

For NR, removing grain, the better the source, the better it can be attacked. When the footage was deinterlaced, and has mosquito noise from compression, the ability to reduce/remove grain is significantly reduced. You're supposed to deinterlace/compress after restoring, not before.

I'm actually more noxious nauseated (thanks Sheldon) watching the footage because the camera is about as steady as a bowl full of jello. I stabilized one of the samples, see attached.

[sanlyn]

I'd echo that it's a bad transfer with many basic errors. I'd guess that you could probably do just as well with a cheap USB capture card and the generic software that comes with it. I'll give a few reasons why I think so:

A) if you suspect that your original source would require restoration and repair or other image mods such as contrast and color correction or applying effects such as fades and dissolves, you cripple your effort at the outset by having the source captured to a lossy final delivery codec. "Final delivery" means that such codecs are not designed for further modification without image degradation. Lossy final delivery formats like MPEG and h.264 also contain compression artifacts and signal loss that don't occur with lossless captures.

Also, final delivery formats are interframe encodes, meaning that even simple cut and join edits entail yet another damaging stage of lossy re-encodes. For simple edits, such encodes require smart-rendering editors to avoid further quality loss. Almost all free editors and some cheap paid editors cannot smart-render h.264 encodes.

B) Your mp4's were captured with an inferior and damaging deinterlace method that dropped alternate fields or frames, and seems to have used an inferior deinterlace method (yes, it does look like yadif) that resulted in many blurry and fuzzed edges. The original Hi8 was interlaced and after field dropping it was encoded as progressive, which actually causes playback problems that look like bad de-interlacing on DVD players. Besides losing 50% of your original color by encoding directly to YV12, you lost half of your original video frames and 50% of the original temporal resolution. None of these losses can be repaired, and edges cannot be smoothed or sharpened to mask the deinterlaced resizing effects. If the person who made these transfers claims that because they are progressive mp4's they are suitable for the internet, they're wrong: 720x480 anamorphic frames can't be used for posting to sites like YouTube or Facebook. No one makes this many newbie mistakes should be called a "pro" and get paid for it.

C) Further, the audio has been encoded with low bitrate AAC audio at 44.1KHz. If you wanted DVD or standard def BluRay for final output, AAC audio at 96Kbps and 44.1 KHz cannot be used. Audio would have to go through another lossy stage of audio resampling and re-encoding for DVD or BluRay.

D) There was apparently no control over captured input levels. On almost every sample, levels are invalid for standard digital video, meaning that levels exceed the range of luminance y=16 to y=235 and chroma UV=16 to UV=240. What this means is that invalid brights and highlights are clipped, and invalid darks or subzero blacks are crushed (i.e, the same thing as clipped). Clipped data is destroyed data -- there is no detail in clipped areas, since all values below or above the clipping value have been converted to the same value, so lost detail cannot be retrieved after clipping. Clipped darks are always zero black and will never be any other color.

Don't confuse the noise in dim, underexposed, mottled, overly dark areas by thinking of it as "grain". Most of the noise in underexposed camera video is sensor noise, not grain. When the signal level is too dark, the signal strength of the camera's residual sensor noise is greater than the signal strength of the incoming image. Sensor noise differs from grain in that grain contains data of various kinds of values but sensor noise contains no usable image information (Sensor noise becomes zero-black and contains no other data). You can do all the filtering or brightening you want, but black sensor noise will always look like black mottling and won't go away.

CMOS noise patterns after brightening, contrast masking, and filtering:


Ann example of processing for gross underexpose is given here, posted just days ago: http://www.digitalfaq.com/forum/vide...html#post55882.


These same videos were seen in other posts a few years ago, in (I think) a different forum. I believe lossless capture was the advice given for best results.

[lordsmurf]

Quote:

Originally Posted by sanlyn

after field dropping it was encoded as progressive, which actually causes playback problems that look like bad de-interlacing on DVD players

Some clarification is needed here. Yadif isn't field dropping. Field dropping is much worse, the "odd" or "even" deinterlace methods. Again, I'm not sure if this was Yadif, or the much earlier adaptive method. The nature of deinterlace does remove a field, but yadif/adaptive take some of each to create a new frames from the 2 former fields. (And of course, both are craptastic compared to QTGMC, which adds in quality anti-aliasing to the newly created frame).

Quote:

C) Further, the audio has been encoded with low bitrate

And like video, lossy audio prevents any further restoration. AAC does allow for reduced bitrates compared to AC3, but compressed is compressed. Also 44.1kHz is missing some of the fidelity found in 48kHz, and that matters for hiss/noise correction. While consumer analog tape video was subpar, the audio was not.

Quote:

levels are invalid for standard digital video, meaning that levels exceed the range of luminance y=16 to y=235 and chroma UV=16 to UV=240. What this means is that invalid brights and highlights are clipped, and invalid darks or subzero blacks are crushed (i.e, the same thing as clipped). Clipped data is destroyed data

I'd actually blame the camera for this. Home movies shot on camcorders are all over the map. I'd be shocked to see a tape with valid levels. And that does happen, but very infrequently. The white balance alone is often wrong (and not just wrong, but very wrong). The lenses of camcorders left much to be desired.

Quote:

Don't confuse the noise in dim, underexposed, mottled, overly dark areas by thinking of it as "grain". Most of the noise in underexposed camera video is sensor noise

Analog didn't have sensors.

[dpalomaki]

Quote:

Analog didn't have sensors.

What were they?

As I understand it, image sensors (typically CCD and later CMOS in handicams) had two noise components: Thermal noise (within the sensors noise) that is random and that under normal exposure (i.e., good light) conditions would be below black level and not visible on screen, but with poor light and AGC it rises above the black level and is visible. Also there can be a grain-like component that is fixed and results from slightly different sensitivity of the individual imaging cells in the sensor.

[sanlyn]

Quote:

Originally Posted by lordsmurf

Quote:

Some clarification is needed here. Yadif isn't field dropping.

When yadif deinterlaces for same-rate output, it drops alternate fields and makes no attempt to create 2 full frames from two interlaced fields. Only one frame is created. Yadif and QTGMC alike create only one frame from each two interlaced fields for same-frame-rate deinterlacing.

[lordsmurf]

Quote:

Originally Posted by dpalomaki

What were they?

I don't remember. I actually looked in a couple of my analog books for the diagrams of how VHS and Video8 cameras were made, for probably 30 minutes without finding it, between writing that and submitting the reply. What I can say is mostly that analog and digital are not the same. Last time I dissected a camcorder was probably the 90s. Those may have CCD, but those formats predate CMOS.

Quote:

Originally Posted by sanlyn

When yadif deinterlaces for same-rate output, it drops alternate fields and makes no attempt to create 2 full frames from two interlaced fields. Only one frame is created. Yadif and QTGMC alike create only one frame from each two interlaced fields for same-frame-rate deinterlacing.

Yadif creates (interpolates) a new frame with data from surrounding fields. I was not aware that Yadif could create 59.94fps until just now, having never used it. But it appears that you must specifically change the mode to 1, and the default mode 0 outputs the input fps. So is it internally creating 59.94 already, and then dropping that? I'm not sure that is the case.

With QTGMC, however, the default is 59.94fps, after creating frames from fields using surrounding data AND advanced processing (anti-alias, NR, etc). Very advanced, very powerful. To match input fps, you must drop a newly created frame.

And just for mentions, to expand on adaptive:
The older adaptive was often a bob with some weak anti-aliasing applied, and sometimes took nearby frames into account. "adaptive" was an overrused term, and described anything not basic drop-frame back in the early 2000s, before the methods like nnedi, Yadif, and QTGMC existed.

[dpalomaki]

Quote:

I actually looked in a couple of my analog books for the diagrams of how VHS and Video8 cameras were made, for probably 30 minutes...

FWIW: I believe the early cameras sold with VCRS (in the late 1970 aned early 1080s) were using vidicons (similar to image orthicons, etc.), and were based on a scanned line analogous to the raster scan of a analog tube TV image. (I had a RCA CC-002 and later a CC-017 back then.)

Development of the CCD in the early 1980s led to its replacing the larger, heavier, more-power-required tubes in video cameras and camcorders. For example, the 1990 vintage Canon Hi8 A1Digital camcorder used 400K pixel 1/2" CCD as the image sensor.

[RyfromNY]

Quote:

Originally Posted by RyfromNY

Sorry for the delayed response. The film was shot between April of 1994 and November of 1995. It was transferred by a service from a master Hi8 to MP4. The original Hi8 was shot on a 1990 Sony Handycam. I'm trying to determine if the level of grit and grain was either a poor transfer by the service, irreversible degradation of the tape itself, or just how a Hi8 from 1994 would look. Basically, if I was to use a higher end service, if I would get better results or if this tape has been screwed from being converted a few times - if I can get anything better. I do notice that in BRIGHT lighting MOST of the grain disappears.

First segment - shot April 1994 in the relatively low light conditions of my parents' bedroom, beginning of tape

Second segment - Shot June 1994 in the sunlight of our backyard

Third segment - Shot July 1994 in the relatively low light conditions of our living room

Fourth - Shot December 1994 in relatively well light conditions at my school

Fifth - Shot November 1995 in very low light conditions in my parents' bedroom.

All from same tape.

Thanks for the responses, so what do you suppose I do next?

I also compared this tape to other tapes and I noticed something - either the aspect ratio is a bit off, or there's very small black vertical bars on the edge of the screen, because, I actually measured it, each side is a little more vertically narrow than on other home movies I have which are also 4:3. Why is this?

Also, can your service guarantee the best possible result?

[lordsmurf]

Quote:

Originally Posted by RyfromNY

Thanks for the responses, so what do you suppose I do next?

A re-capture is honestly required, if you want a quality transfer. Who does that is up to you, either DIY or another service.

Quote:

I also compared this tape to other tapes and I noticed something - either the aspect ratio is a bit off, or there's very small black vertical bars on the edge of the screen, because, I actually measured it, each side is a little more vertically narrow than on other home movies I have which are also 4:3. Why is this?

It really depends on the tape and sources. Not all NTSC signals contained image data, and some were actually 704 palette matted to 720. --- But another option is that the service you used jacked up the video file, and seeing what I've seen so far, that cannot be eliminated as a possibility. I'd have to see the original tape.

Quote:

Also, can your service guarantee the best possible result?

I'd say that's a yes. I'll let our site guides and staff forum posts speak for themselves, in this regard. The only real time that "best possible" isn't done is because the budget doesn't allow for it. We get some really tough and ugly restoration projects, not just those needing good transfers. And when something is not possible, we can at least give you an exact reason why it's the case.