Digital Media Group
<dmg at mir.com>
(You can skip directly to the Handy Reference Table. The following discussion is actually a simplification of the truth, but it captures the important details. See the references for even more details.)
In computer graphics, pixels are square. In video, pixels are rectangular. This is equivalent to saying that video has different spatial sampling in the horizontal and vertical directions.
A square pixel has a pixel aspect ratio (or sample aspect ratio [SAR]) of 1:1. A non-square pixel has an aspect ratio W:H where W and H are not equal.
Do not confuse this with the display aspect ratio [DAR], which describes the shape of the physically displayed image. This describes the shape of the frame as measured with a ruler.
The other two parameters describing video frame are the width and height measured in pixels. These parameters are related to each other as follows:
width DAR -------- = ----- height SAR
The industry standards (i.e. as practiced) for digital video sampling are derived from ITU-R BT.601 (also known as "CCIR-601" or "Rec.601"), and presume a horizontal sampling rate of 13.5 MHz and an active line of 720 samples. The pixel aspect ratio for 525-line video (NTSC) is 10:11. The pixel aspect ratio for 625-line video (PAL) is 59:54. Exactly. (Well, not exactly; see "What is square anyway?" below.)
An image with a standard display aspect ratio of 4:3 (i.e. TV) with 480 lines (i.e. NTSC) would thus have:
An image with a standard display aspect ratio of 4:3 (i.e. TV) with 576 lines (i.e. PAL) would thus have:
Standard (non-widescreen) provides a frame size of 720x480 (DV-NTSC) or
720x576 (DV-PAL) using appropriate Rec.601 pixel aspect ratios. Thus,
according to the previous calculations, these frames contain a visual
image which is slightly wider than TV's 4:3 display aspect ratio.
To be displayed properly in a 4:3 screen or window, the frames should be
cropped by 8 pixels on each side.
Widescreen Format
Standard TV has a 4:3 aspect ratio. The new widescreen format (HDTV, anamorphic DVD's, wide DV, etc.) has a 16:9 display aspect ratio. Frames are encoded at the same size, but with an appropriately scaled pixel aspect ratio:
An NTSC VCD contains an MPEG-1 video stream with 352x240 frames. The NTSC-VCD pixel aspect ratio is just Rec.601's standard 10:11. The frames implicitly are downsampled from a source containing 704x480 pixels. This is equivalent to a truly 4:3 source containing 640x480 square pixels.
Similarly, a PAL VCD contains on MPEG-1 video stream with 352x288 frames. The PAL-VCD pixel aspect ratio is just Rec.601's standard 59:54. The frames implicitly are downsampled from a source containing 704x576 pixels. This is equivalent to a source containing approximately 768x576 square pixels.
DV source material (with 720x480 or 720x576 sized frames) must be cropped by 8 pixels on each side (16 total) before scaling and encoding for VCD.
Exact 4:3 source material with Rec.601 pixels (i.e. 704x480 or 704x576 sized frames) or square pixels (i.e. 640x480 or 768x576 sized frames) does not require any cropping or padding before scaling and encoding for VCD.
The actual MPEG-1 specification specifies the following pixel aspect ratios:These numbers are self-consistent, but otherwise rather contrived. They should have just said "10:11 for CCIR Rec.601, 525-line" and "59:54 for CCIR Rec.601, 625-line" since that is apparently what they meant and that is how they are used.
- 1.0950 for "CCIR Rec. 601, 525-line" images with frames of "711x487 at 4:3"
- 0.9157 for "CCIR Rec. 601, 625-line" images with frames of "702x575 at 4:3"
SVCD's are, like everything in the video world, just a little bit different.
An NTSC SVCD contains on MPEG-2 video stream with 480x480 frames. The frames implicitly are scaled from a source containing 720x480 Rec.601 (10:11) pixels. This is equivalent to a source containing approximately 654.5x480 square pixels. One could even say that the NTSC-SVCD pixel aspect ratio is:
10 720 480 10 * 3 15
Rp = ---- * ----- * ----- = -------- = ----
11 480 480 11 * 2 11
Similarly, a PAL SVCD contains on MPEG-2 video stream with 480x576 frames. The frames implicitly are scaled from a source containing 720x576 Rec.601 (59:54) pixels. This is equivalent to a source containing approximately 786.67x576 square pixels. One could even say that the PAL-SVCD pixel aspect ratio is:
59 720 576 59 * 3 59
Rp = ---- * ----- * ----- = -------- = ----
54 576 480 54 * 2 36
DV source material (with 720x480 or 720x576 sized frames) can be directly encoded to SVCD --- with 2:3 scaling, of course, but no padding or cropping.
Exact 4:3 source material (with Rec.601 pixels, i.e. 704x480 or 704x576 sized frames) needs to be padded to a width of 720 before scaling and encoding for SVCD.
Square-pixel 4:3 source material (640x480 or 768x576 sized frames) likewise needs to be padded (to widths of 654 and 786 respectively) before scaling and encoding for SVCD.
According to the actual MPEG-2 specification, the video stream sequence header specifies the intended display aspect ratio rather than the pixel aspect ratios, from one of these choices:(Oh, and you can also specify "1:1 pixel aspect ratio".) This is supposed to be an improvement over MPEG-1's contrived pixel aspect ratio descriptions. It's not, because in practice, MPEG-2 sources are 720x480 or 720x576 with CCIR-601 pixels, which are slightly wider than 4:3 or 16:9. So, as before, the actual numbers in the specification don't mean very much.
- 4:3
- 16:9
- 2.21:1
Aspect ratios are in terms of "WIDTH:HEIGHT". Frame sizes are in terms of "WIDTHxHEIGHT".
| system | source/dest | format | pixel aspect ratio | frame size |
|---|---|---|---|---|
| 525-line, NTSC | framegrabber | 4:3 | 10:11 | 704x480, 720x480, 352x240 |
| DV | 4:3 | 10:11 | 720x480 | |
| 16:9 | 40:33 | 720x480 | ||
| graphics | 4:3 | 1:1 (square) | 640x480 | |
| 16:9 | 1:1 (square) | 854x480 | ||
| VCD | 4:3 | 10:11 | 352x240, 704x480 (stills) | |
| CVD | 4:3 | 20:11 | 352x480 | |
| SVCD | 4:3 | 15:11 | 480x480 | |
| 10:11 | 704x480 (stills) | |||
| 16:9 | 20:11 | 480x480 | ||
| 625-line, PAL | framegrabber | 4:3 | 59:54 | 704x576, 720x576, 352x576 |
| DV | 4:3 | 59:54 | 720x576 | |
| 16:9 | 118:81 | 720x576 | ||
| graphics | 4:3 | 1:1 (square) | 768x576 | |
| 16:9 | 1:1 (square) | 1024x576 | ||
| VCD | 4:3 | 59:54 | 352x288, 704x576 (stills) | |
| CVD | 4:3 | 59:27 | 352x576 | |
| SVCD | 4:3 | 59:36 | 480x576 | |
| 59:54 | 704x576 (stills) | |||
| 16:9 | 59:27 | 480x576 |
Note that one typically tries to keep a constant frame height, by scaling the width as needed. With interlaced frames, scaling between scanlines becomes a complicated operation.
Note also that all the non-square pixel aspect ratios are
derived in some way from the CCIR-Rec.601 ratios (via some combination
of 3:4 and/or 2:3 to account for widescreen and/or SVCD).
What is "square" anyway?
Just a note about the industry standards: With the sampling rates chosen by the Industry, "Industry Standard square pixels" are not actually square! According to Aho (see References), they have an aspect ratio of 768:767, which is very close to 1:1, but which is not truly exactly 1:1. To arrive at the "true" physical pixel aspect ratios of the various formats, all the numbers listed above need to be multiplied by 768:767.
However, all the conversions between formats --- e.g. any scaling to convert from one to another --- are referenced to the "Industry Standard square pixel". Unless you need to make physical measurements off of your screen, you will want to use the numbers above and live under the illusion that the Standard square pixel is truly a square pixel.
|
|
maximum impact research, Digital Media Group <dmg at mir.com> |
Last modified: Wed Mar 3 12:47:41 EST 2004 |
©2003 Matthew Marjanovic.
This material may not be republished in any form
without express written consent of the author.