SAR, DAR, PAR: why 720x480 is not 3:2 - pixel aspect ratio explained
Here is a puzzle. You have a video file that is 720x480. Divide width by height and you get 1.5 - a 3:2 ratio. So the picture is 3:2, right? Except the very same 720x480 grid can display as 4:3 in one file and 16:9 in another. Same pixel count, same resolution string, two completely different shapes on screen. The naive width-over-height math is simply wrong, and most people never find out why.
The reason is one assumption that almost everyone makes without noticing: that pixels are square. For a lot of standard-definition and broadcast video, they are not.
Pixels are not always square
The mental model that makes this click is to think of the stored frame as a wall of bricks. Each pixel is a brick. If your bricks are perfect squares, the wall has the same shape as the grid - width over height tells you everything. But if the bricks are taller than they are wide, the finished wall ends up narrower than the grid suggests. If the bricks are wider than tall, the wall comes out wider.
That brick's own width-to-height ratio is the Pixel Aspect Ratio (PAR). When PAR is 1:1, pixels are square and the easy formula works. When PAR is anything else, the stored grid and the displayed shape part ways.
The one formula that ties it together
Everything in this article comes down to a single relationship:
DAR = Storage Aspect Ratio x Pixel Aspect RatioWhere the Storage Aspect Ratio is just the coded grid, width / height, and the Display Aspect Ratio (DAR) is the shape the viewer actually sees. Written out with the pixel ratio as PARw:PARh:
DAR = (W / H) x (PARw / PARh) = (W x PARw) / (H x PARh)In plain words: take the stored width, multiply by how wide each pixel is, and that gives you the effective display width. The picture you see is effective_width : H.
Run our mystery 720x480 grid through it. The storage ratio is 720/480 = 3:2. Now apply the pixel ratio:
- With PAR
8:9(pixels slightly taller than wide):(3/2) x (8/9) = 4:3. - With PAR
32:27(pixels noticeably wider than tall):(3/2) x (32/27) = 16:9.
One grid, two pixel shapes, two display ratios. The 3:2 you computed from the raw resolution was never meant to be displayed - it is the shape of the storage, not the picture.
Rearranging the same equation gives the two lookups you will reach for in practice: PAR = DAR / (W/H) when you know the display shape and want the pixel ratio, and W/H = DAR / PAR when you are sizing a grid.
The acronym trap
If the math were the whole story this would be a short article. The real difficulty is that the field is buried under three-letter acronyms that collide with each other - sometimes meaning the exact opposite thing.
PAR usually means Pixel Aspect Ratio - the shape of one pixel. But plenty of documentation uses PAR for Picture Aspect Ratio, which is the display shape, i.e. the DAR. Same three letters, two quantities on opposite sides of the formula. A rough rule that holds most of the time: "PAR 16:9" almost certainly means Picture (the display), while "PAR 8:9" or "PAR 10:11" means Pixel (the brick).
SAR is worse. In the H.264/AVC and HEVC bitstream, the VUI field called Sample Aspect Ratio is the pixel shape - SAR there is exactly the PAR we have been computing with. But other applications use SAR for Storage Aspect Ratio, coded_width : coded_height - the other factor entirely. Amped FIVE's documentation (opens in new tab) explicitly distinguishes "SAR (Storage Aspect Ratio)" from "SAR (Sample Aspect Ratio)" and shows both in its file information. FFmpeg uses the latter meaning, so probing a 720x480 anamorphic file prints something like:
SAR 32:27 DAR 16:9Here FFmpeg's "SAR" is the sample/pixel ratio, and the formula holds: (720/480) x (32/27) = 16:9. The lesson is not to memorize which tool means which - it is to always check what a given "SAR" refers to.
DAR is the one acronym you can mostly trust: Display Aspect Ratio, the shape on screen, whether that is 4:3, 16:9, or a 2.21:1 cinema frame.
A gallery of real formats
Here is the formula doing its job across the formats you will actually encounter. Every row satisfies DAR = (W/H) x PAR.
| Stored grid | PAR | DAR | Notes |
|---|---|---|---|
| 720x480 NTSC | 8:9 | 4:3 | full-720 width |
| 720x480 NTSC | 32:27 | 16:9 | anamorphic widescreen, full-720 |
| 704x480 NTSC | 10:11 | 4:3 | BT.601 active area |
| 704x480 NTSC | 40:33 | 16:9 | BT.601 active area |
| 720x576 PAL | 16:15 | 4:3 | full-720 |
| 720x576 PAL | 64:45 | 16:9 | anamorphic, full-720 |
| 704x576 PAL | 12:11 | 4:3 | BT.601 active area |
| 704x576 PAL | 16:11 | 16:9 | BT.601 active area |
| 1440x1080 HDV/HDCAM | 4:3 | 16:9 | tall pixels, effective width 1920 |
| 1280x1080 DVCPRO HD | 3:2 | 16:9 | effective width 1920 |
| 1920x1080 HD | 1:1 | 16:9 | square pixels - the easy case |
Two traps hide in this table.
The first is the 720-versus-704 mismatch. Notice that 10:11 produces 4:3 only at 704 width. People remember "10:11 is NTSC 4:3" and then apply it to a full 720 grid, where (720/480) x (10/11) = 15:11, about 1.36 - not 4:3 at all. The PAR you pick has to match the width you pick. Borrowing a 704-active-area pixel ratio for a 720 grid (or vice versa) is one of the most common real-world aspect-ratio bugs.
The second is more conceptual. Look at the bottom three rows. 1920x1080 has square pixels, so the naive width/height formula "just works" and gives 16:9 - and that is exactly what fools people. They generalize from HD, assume the formula always works, and then get burned by 1440x1080 (HDV) or 1280x1080 (DVCPRO HD), which store a true 16:9 picture in a narrower grid using wide pixels. The 1920 case and the 1440 case are the cleanest side-by-side demonstration of why you cannot trust resolution alone.
The circle test
The quickest way to see a wrong pixel aspect ratio is to draw a circle. A true circle stored in a frame with non-square pixels does not look like a circle in the raw grid - it looks like an ellipse. Apply the correct PAR and it snaps back to round.
This is exactly what anamorphic DVD does: 16:9 content is squeezed horizontally into a 4:3-shaped 720x480 (or 720x576) grid with wide pixels, then stretched back out on playback. The word is overloaded, though. In the DVD world it means non-square-pixel 16:9. In cinema it means an optical lens that squeezes a wide field onto the film and is un-squeezed in projection to a Scope-class widescreen frame. Both are "anamorphic," and they are not the same mechanism.
Where the values live in the bitstream
So where does a player find the pixel ratio? In a different place in every format - and each format encodes a slightly different idea.
MPEG-2 carries aspect_ratio_information in the sequence header, a 4-bit code. The quirk is that it mixes two kinds of ratio in one field: code 1 means a 1:1 square sample (a pixel ratio), while codes 2, 3, and 4 mean 4:3, 16:9, and 2.21:1 - which are display ratios. So MPEG-2 mostly stores the DAR directly, even when the coded size is a 720x480 (3:2) grid.
H.264/AVC and HEVC put aspect_ratio_idc in the SPS, inside the VUI parameters. This 8-bit index selects a Sample (= pixel) Aspect Ratio from a table - for example 1 = 1:1, 2 = 12:11, 3 = 10:11, 4 = 16:11, 5 = 40:33, 14 = 4:3, 15 = 3:2, 16 = 2:1. The special value 255 is Extended_SAR, which means explicit 16-bit sar_width and sar_height fields follow. Because this is a pixel ratio, you still have to compute the DAR yourself as (coded_w / coded_h) x SAR - and apply any cropping window first.
MP4 and QuickTime store the pixel ratio in the pasp box inside the visual sample entry, as hSpacing and vSpacing (width:height). The sample entry's own width and height are the coded grid. Meanwhile the tkhd track header has its own width and height - 16.16 fixed-point display dimensions - which can legitimately disagree with the coded size (coded 1440x1080 but tkhd saying 1920, for instance). That means up to three places in one file can carry size information that must be reconciled. If there is no pasp box at all, you assume square pixels.
There is also a historical twist. MPEG-1 used the same 4-bit slot for a field called pel_aspect_ratio, and it encoded the pixel ratio as height:width - vertical over horizontal. Every modern convention (H.264 SAR, MP4 pasp, FFmpeg PAR) expresses the pixel ratio as width:height. So the ordering flipped between generations, and then MPEG-2 repurposed the field entirely from a pixel-ratio table into a display-ratio table. If an old decoder and a modern one disagree about which way a ratio points, this inversion is often why.
See it yourself in VTCLab Media Analyzer
The fastest way to internalize all of this is to open real files and watch the values line up. VTCLab Media Analyzer exposes each of these fields where the format stores it:
- MPEG-2 - open the stream, find the sequence header, and read
aspect_ratio_information. You will see it hold a display code (2 for 4:3, 3 for 16:9) even when the coded size is 720x480, a 3:2 grid. - H.264 / HEVC - drill into the SPS, then VUI, and read
aspect_ratio_idc. If it is 255, read the explicitsar_widthandsar_height. Then computeDAR = (coded_w / coded_h) x SARfor yourself. - MP4 - inspect the
paspbox (hSpacing/vSpacing) and contrast it with thetkhddisplay width and height. This is where the container-versus-stream and coded-versus-display discrepancies become visible in one view.
What makes the comparison worthwhile is that the three formats encode the same concept three different ways: MPEG-2 stores the display ratio, H.264 stores the pixel ratio, and MP4 stores a pixel ratio in a box plus a display size in tkhd. Open one of each and the same equation holds every time:
DAR = Storage Aspect Ratio x Pixel Aspect RatioOnce you stop trusting width / height and start asking what shape the pixels are, 720x480 being both 4:3 and 16:9 stops being a paradox - and becomes just arithmetic.
Комментарии