There is a lot of advice on this forum about recommended capture pixel format for color. Generally 4:4:4 is considered overkill and 4:2:2 is considered the best we can do from an NTSC signal (8mm in my case).
I tend to question everything, so I have attempted a calculation, but not sure if scan lines per frame should be 480 or 525 - used 480:
Edit: spreadsheet had errors - see below.
Spreadsheet updated. See Post Below.
+ Reply to Thread
Results 1 to 30 of 82
Last edited by GrouseHiker; 21st May 2020 at 12:39. Reason: Corrected spreadsheet.
Consumer video tape has far less chroma resolution than theoretically possible by the NTSC spec. Even 4:1:1's 176 samples (across the 704 pixels of the active picture) is sufficient for the chroma. Take one of your 4:2:2 caps and try this:
Interleave(last, MergeChroma(last, BilinearResize(width/4,height).Spline36Resize(width,height)))
Last edited by jagabo; 21st May 2020 at 10:02.
[Attachment 53386 - Click to enlarge]
Trying to figure out your statement:
Even 4:1:1's 176 samples (across the 704 pixels of the active picture) is sufficient for the chroma.
[Attachment 53590 - Click to enlarge]
If my assumptions are correct on color content of the original signal (i.e., based on color carrier modulation frequency), color content lies a bit below 4:2:2. Reality is probably different, as I'm sure will be demonstrated by your code.
A question: Is my use of 480 scan lines correct?
The chart is updated below:
Last edited by GrouseHiker; 29th May 2020 at 12:00. Reason: updated chart
In a perfect theoretical NTSC analog signal, yes 4:2:2 is sufficient for capture. But you're dealing with a consumer analog tape. The chroma that's stored on the tape has far less resolution than the NTSC spec allows for.
Yes, 480 scan lines is correct. The actual active picture is 485 scan lines -- the topmost scanline containing actual picture on the second half of the line, the bottom most scanline only the first half of the line, and 484 full scan lines in between. When captured those half lines become full lines, for a total of 486 lines. But even those 486 lines usually don't all contain real picture information. Some lines are used for closed captions and other flags. From consumer tape there is head switching noise at the bottom of the frame. So typically only 480 scan lines are captured. 480 has the additional benefit that it's mod 32-- so mod 32 and mod 16 codecs can deal with it.
The remainder of the 525 scan lines that define the full NTSC signal are used for sync in analog TVs.
Someone posted a full 525 line cap from a VHS tape a while back:
Last edited by jagabo; 21st May 2020 at 13:07.
Last edited by dellsam34; 21st May 2020 at 15:48.
I'm getting way beyond my knowledge level, but comparing to the NTSC standard, the bandwidth contained on tape appears quite a bit less:
[Attachment 53391 - Click to enlarge]
If bandwidth is the proper comparison, the Video8 has (based on the Sencore sheet) 743 kHz chroma while NTSC standard shows 2 MHz chroma. Is that apples-to-apples?
I'm now wondering if my calculation is proper for NTSC... should I use bandwidth instead of subcarrier frequency? If yes, that would really bring the numbers down. Maybe bandwidth relates to color bit depth and subcarrier frequency relates to sampling interval?
Another thing that's got me stumped is the pixel color sampling is based on two, stacked, 4-pixel groups. If a scan line is one pixel high, where does the second, 4-pixel group come from?
483 scan lines per frame + 42 lines vertical blanking = 525 scan lines.
or 262.5 lines per field * 2 fields = 525 scan lines.
Are you dealing with a 4:3 aspect ratio throughout?
I used to know this stuff by heart; now I have to sit down and figure it out.
Keep in mind that your analog 8mm NTSC tape has a phase-modulated color subcarrier (sine wave) so you have to be cautious when doing calculations for digital capture.
That line rate doesn't look right, either. Should be more like 15,734 Hz or 3.579545 × 2/455 MHz. You also have to factor in horizontal blanking (10.9 us).
I get 188 cycles of 3.579545 subcarrier per active H (52.65 us, excluding H blanking). Does that sound right?
Last edited by chris319; 21st May 2020 at 19:05.
A video standard (or video system) describes the properties of a particular video signal used by a device (such as a camcorder). For example, the NTSC standard used for broadcasts and baseband video in the Americas, Japan, Korea, and Taiwan, defines a frame rate of 29.97 fps (frames per second), 525 horizontal lines, and a color subcarrier at 3.58MHz. Camcorders sold in these regions must be able to generate a signal that complies with this standard.
This is not to be confused with a format, which defines how the video is encoded or recorded to a medium (this typically has regional standards of its own); and in the case of tape formats, what kind of tape to use, and cassette dimensions. Physical properties of a format are usually universal.
Tape formats used in camcorders usually have their own unique variants of their native standard.
The various questions/comment have led me to a different way of looking at this:
The 8mm player is sending out a signal intended for TV with a resolution of ??? x 480-ish. Without knowing the best-possible horizontal resolution from 8mm, it seems the answer can't be calculated... back to trial and error and looking at pixels.
Another complication is that you haven't separated chroma and luma. You're sampling them together. Your 8mm analog original doesn't know about subsampling.
For H sampling the minimum is 4 x 188 = 752 samples (this is for the visible part of the scan line and does not include horizontal blanking).
You could also use 768 x 486 which is compliant with SMPTE 259M. It will capture a few pixels of H blanking and a few lines of V blanking, no big deal.
You also have to consider whether your digitized video will play with the correct aspect ration.
858 x 483 will capture all of your scan lines and make a 16:9 picture.
In 2020, where are you going to find a 4:3 display? They're all 16:9 now.
Last edited by chris319; 22nd May 2020 at 07:04.
On top of that the resolution of the camera or broadcast was even with less resolution....
Accordig to that Sencore tech tips paper... Hi8 was the best analog tape format, even better than U-matic.... shame about those crappy optical sensor chips used in those cameras. though Canon had some good models.. L-2, EX1, only for the semi pro user..
Last edited by Eric-jan; 22nd May 2020 at 07:53.
Video Format.............................. Horizontal Resolution (resolution width)
Standard VHS............................. 210 Vertical "Lines"
Hi8............................................... 400 Lines
DV...............................................5 00 Lines
DVD............................................540 Lines(?) [some actual DIGITAL sizes: 720(w)x480(h), 704(w)x480 or 352(w)x480 ]
...VHS, 8mm, and 3/4-inch U-Matic VCRs can record about 240 lines of resolution. These, incidentally, are black-and-white lines or luminance resolution. Color sharpness is much lower but is not very noticeable to the eye. Since luminance resolution is responsible for most of what we preceive as a sharp picture, we will talk about luminance resolution. SVHS and Hi8 VCRs can record and play back 400 lines of luminance resolution. This higher resolution is only seen if you are using the Y/C connectors. If you use the composite outputs of these machines, the resolution is generally reduced to 330 lines. Broadcast television also uses 330 lines as its maximum picture resolution...
This statement from wikipedia.org/wiki/Chrominance, indicates Chroma Subsampling is related to bandwidth, not the carrier frequency:
Separating RGB color signals into luma and chrominance allows the bandwidth of each to be determined separately. Typically, the chrominance bandwidth is reduced in analog composite video by reducing the bandwidth of a modulated color subcarrier, and in digital systems by chroma subsampling.
I don't know what are you trying to achieve here? The wheel has already been invented. What you can do is get a pro capture device that adhere to the SMPTE 259M standards, Capture at 720x486 4:2:2 8Bit, then crop to 704x480. the extra 6 lines will get rid of the head switch and V blanking pulses and the padded 16 pixels (8 on each side) that was added in the ITU-Rec 601 (CCIR-601 or D1) standards and adopted by every capture card out there will be removed. You will be left with about 95% of clean video as the picture shows bellow (no aspect ratio flag set yet).
From there if you want to reduce size encode to H.264 4:2:0 8Bit, give each pixel a SAR of 10/11 and keep a decent audio quality @ 192Kbps.
Last edited by dellsam34; 22nd May 2020 at 14:55.
While I waited for a replacement player to come in, I was trying to prove to myself 4:2:2 8-bit is the best we can do (or maybe overkill). I have no problem capturing 4:4:4 10-bit, but I'm thoroughly convinced this would be WAY overkill for Video8 capture. I have the Magewell Pro Capture HDMI card, and the eBay non-digital camera came in late yesterday. Guess I'll be riding that wheel that was already invented. I'll include a comparison with this new camera back in my thread comparing Composite, S-Video, and DV... hopefully it will be better.
I will have to research the answer to your 4 x 188 question.
The way I understand it, capturing at 858 x 483 and displaying at 16:9 will simply make everything w i d e.
I tried making a video of a test pattern with a circle on it and had to use 858 x 483 so the circle would come out looking round on my 16:9 monitor.
You can simulate some of this. Do you know how to program graphics or write an ffmpeg script?
Last edited by chris319; 22nd May 2020 at 12:30.
You would not gain anything from 4:4:4 at least visual, 10bit comes in handy if you do a lot restoration with filters otherwise 8 bit should be fine, shame that most script type application work in 8 bit only. S-Video is the best output and closest to the tape. it has been compared already. What you need is a good camcorder with line TBC.
Vdub you have the option to crop as many pixels as you want on each side as long as the total removed pixels is 16. Once you have only active video then you have to correct the aspect ratio by setting the SAR to 10/11 because capture cards used more resolution horizontally than what the 4:3 aspect ratio calls for, which is good for better details.
As I said, I made a test at 720 or 768 (I forget which) using a test pattern with a perfect circle, and the circle was distorted, i.e. not perfectly round on my monitor which is set for 16:9.
Maybe there was a flaw in the way I generated the file (ffmpeg) but that's what I got. I was trying to emulate the O.P.'s 4:3 original image.
I'm sure the O.P. doesn't have a test pattern on hi 8 tape.
[Attachment 53450 - Click to enlarge]
ImageSource("frame.jpg") ConvertToYV24() StackVertical(StackHorizontal(last, GreyScale().SubTitle("Y")), StackHorizontal(Tweak(sat=3.0).UtoY().Subtitle("U"), Tweak(sat=3.0).VtoY().Subtitle("V")))
A simlar plot from one of your YUY2 video:
[Attachment 53453 - Click to enlarge]
And a true YV24 video (downscaled from 4K to SD)
[Attachment 53455 - Click to enlarge]
Last edited by jagabo; 22nd May 2020 at 20:49.
Do you have a test pattern or something with a perfect circle that you can shoot with your camcorder?
Last edited by GrouseHiker; 23rd May 2020 at 19:15. Reason: Removed chart based on dellsam34 comments below... working on corrections.
Last edited by GrouseHiker; 22nd May 2020 at 22:13.
I think dellsam has a camcorder he can use to shoot some 4:3 footage of a round object for a test.
To do it right you need to measure the on-screen dimensions of a round object rather than eyeball it, so his test is potentially very useful.
For both captured 720x480 and cropped 704x480 the PAR is the same which is 10/11. With 704 the DAR is exactly 4:3 but for 720 it is slightly wider (exactly 16 pixels wider), but the active 704 of the 720 is exactly 4:3. Also 486 is not a legal resolution in the SDTV standard, it is a transitional step after capturing and it should be cropped to 480 immediately, The same thing should apply to 720 and should be cropped to 704 right after capturing but because a lot of people are confused between 720 DVD anamorphic widescreen which it contains all 720 active pixels with DAR of 16:9 and the 720 capture (704 active + 16 padded pixels) with a DAR of 4:3, they tend to keep it that way, but they are completely two different things.
If you want to make 16:9 with 1:1 PAR (square pixel) you will have to crop to 704 and convert the 704 to 640 and add approx 213 dummy pixels horizontally, If you want 16:9 with the same 10:11 PAR you will have to add approx 235 dummy pixels to 704 or 219 to 720.
Last edited by dellsam34; 23rd May 2020 at 02:29.