Square pixels vs Interpreted pixels

[Ron Hunter]

I am in the midst of creating a drivelapse using photos taken with my DSLR. I understand that DSLR images contain square pixels.

During video editing I remembered watching a Lynda timelapse tut where the presenter, within his video editing program (PreEl), interpreted his photos as "DV/D1 NTSC (0.9091)". That got me thinking "What's the difference in the final movie file between square pixels and interpreted pixels?", so I performed an experiment. I rendered one timeline with square pixel photos, another timeline with interpreted photos, and placed both output files on top of one another. Both timelines had the identical number of images and were output as 24fps files. Here is a snapshot of the result:

Capture.JPG

I think both of them look pretty good, but it is interesting how:
- Square pixels seem to pull the field of view into the viewer. Things look closer, but some visual information on left and right is lost.
- Interpreted pixels do the opposite; they push the field of view away from the viewer. Things look farther away, but there is more information on left and right sides of screen.

Before I directly compared them I wasn't aware of a difference. When I watched the square pixel video I thought, "That looks good." Then when I watched the intepreted video I thought, "That looks good too." Only when I put them on top of one another could I tell the difference.

Would someone kindly explain the square vs interpreted pixel thing to me?

[Steve Grisetti]

Simply put, you squeezed your video in 10%. The interpolated video is only 90% as wide as the square pixel video.

I'm not sure what advantage there would be to it, despite what you're seeing in your comparison. In fact, in some situations, it would make your picture look too tall and skinny. Not much, but a little. If you shot video of something like a perfect circle or perfect square it would be more obvious.

So I'm not sure what you saw in the lynda course or in what context, but I know of no reason it would be preferable to turn square pixels into .9 pixels.

[Bob]

A few points:

First, if you are using HD, it's all square pixels. Non-square pixels are used in SD.

Second, If you import a square pixel photo into a SD project, PrE knows the pixel aspect ratio (par) of the photo is square. PrE will automatically handle the pixel aspect ratio differences and resample the image to non-square pixels when you place the photo on the timeline. A round circle in the square pixel photo will remain round in the non-square timeline.

Third, if you use interpret footage to change the pixel aspect ratio, it's not doing what you think it's doing. It's not converting the square pixels to rectangular pixels. PrE knows the image pixels are square and will resample them when they are placed on the timeline. What you are telling PrE to do is conform the image by applying a scaling factor before resampling. A round circle in the square pixel photo will become oval in the non-square timeline. If you specify a par of .9, the circle will become squeezed horizontally. If you specify 1.2, it will become stretched horizontally.

[Chris B]

First, if you are using HD, it's all square pixels. Non-square pixels are used in SD.

The exception that proves the rule is of course HDV 1080i footage which is 1440x1080 resolution and 16:9 aspect.

[Bob]

You're right, I should have either included HDV as an exception in my comment or left that comment out entirely.

The key items are points 2 and 3. And, that's true for any non-square pixel project.