Carl Camera

Video Encoding Demystified

A canonical problem you see people make is: I've got a 720x480 source and I want to make it smaller. Then I'll make it shorter so I'll make it 240 tall -- that makes sense -- and then I'll make it thinner so I'll make it 360 wide -- divided by two -- and that's always wrong. Because your image is really either four by three or sixteen by nine, 720x480 is a non-square pixel video frame. So if it's four by three, 320x240 is a good size. If it's sixteen by nine, then 432x240 is a good size. But 360x240 is never the right size for any video.

I've been slowly working my way through the Mix08 videos, which on the whole have been great. Today I watched Ben Waggoner's presentation Encoding Video for Microsoft Silverlight Delivery Scenarios and I was puzzled, then surprised and amazed over this slide...

  • Match aspect ratio of cropped source
    • 4:3 720x480 to quarter size
      • Right: 320x240
      • Wrong: 360x240
    • 16:9 720x480 to quarter size
      • Right: 432x240
      • Wrong: 360x240

... and Ben's accompanying comments (@55:20) ...

I don't encode video often, but I've certainly made this mistake. Ben's presentation also contains this tip for folks who create screencasts: (@39:30) Turn ClearType off when creating the screencast. Cleartype smooths out screen text on LCDs by generating different color pixels around characters. The encoder will detect and attempt to faithfully render these pixels. Ben's advice is that your output video will be crisper if your original source video starts with crisp pixels. I can't argue that point either.

Ben's presentation provides explanations for several heretofore (for me at least) cryptic settings found in most modern nonlinear video rendering programs. The Microsoft Expression Encoder product is showcased prominently, but even if you're dealing with Final Cut Pro or Premiere, there's good information in Ben's presentation.