What Are Non-Square Pixels? Anamorphic Video Explained

Open an old camcorder clip or a film-style recording in the wrong program and everyone in it looks a little too thin — or too wide. The picture is fine. It was stored with pixels that are not square, and the program forgot to stretch it.

Storage size is not display size

A video has two shapes:

• Storage shape: how many pixels wide and high the frames actually are.
• Display shape: how wide and high the picture should appear.

When those differ, the file carries a sample aspect ratio (SAR) — the shape of one pixel. A SAR of 1:1 means square pixels, the normal case today. Anything else means each pixel is meant to be drawn wider or narrower than it is tall.

Display shape = storage shape × SAR. That is all there is to it.

Where non-square pixels come from

• DV and DVD video. Standard-definition video was 720×480 (NTSC) or 720×576 (PAL) whether the picture was 4:3 or 16:9. Widescreen was the same pixel grid, stretched on playback — a SAR such as 32:27 or 40:33 for NTSC widescreen.
• HDV and some broadcast HD. 1440×1080 frames with a 4:3 SAR, displayed as 1920×1080.
• Anamorphic lenses. These squeeze a wide scene horizontally onto the sensor (commonly by 1.33× or 2×), and the recording is flagged to be stretched back out.

Why some tools get it wrong

A player, editor or analysis tool that ignores the SAR draws one stored pixel as one screen pixel. The result is the squeezed storage shape: thin faces, oval wheels, a 16:9 shot displayed as a narrow 4:3 picture.

The video quality checker reports it when it finds it: “The pixels are not square (sample aspect 32:27).” Not as a fault — an anamorphic master is a perfectly normal file — but as the explanation for why one program shows the right shape and another does not.

Converting to square pixels

For delivery to the web, phones and most platforms, square pixels are the safe choice. Resize the width by the SAR and mark the result square:

ffmpeg -i input.mov -vf "scale=iw*sar:ih,setsar=1" -c:v libx264 -crf 18 -c:a copy output.mp4

scale=iw*sar:ih stretches the width to the display shape; setsar=1 labels the result square, so nothing downstream stretches it a second time. That double stretch — a file already widened but still carrying its old SAR — is the other mistake to watch for: everything comes out too wide.

Comparing an anamorphic master to its export

Comparing a non-square master against a square-pixel export of the same footage compares two different frame shapes. Convert the master as above first, then compare the two in DiffALL’s video comparison — the differences left are the ones the export actually introduced.