View Full Version : A resolution experiment someone should try...

09-28-2010, 03:22 PM
1. Record a few seconds of a scene with some high frequency detail in it. A resolution chart would be best.

Stop the record.

2. Without moving the camera, hit the digital zoom button twice for 10x magnification.

Compare the detail in the .mov VS. the detail in the digitally zoomed camera LCD/HDMI out. Clearly the digitally-zoomed in image contains more detail than if one was to take the un-zoomed .mov and blow it up in post.

Seems to me this camera might be capable of much sharper footage than what we are currently getting. Check out the side -by-side below.

One image is a picture of the camera LCD, in movie mode, zoomed in 10x.
The other is 2x magnification of .mov file in VLC.

Both taken with a 50mm lens from the exact same position, only difference is the glass was moved in one of the pictures.

Way more detail in the in-camera zoommed image. You can make out some of the letters and numbers on the paper. I took the pic of the LCD with my blackberry, it's actually quite sharper in person. Interesting, no?


http://www.dvxuser.com/V6/%3Ctable%20style=%22width:auto;%22%3E%3Ctr%3E%3Ctd %3E%3Ca%20href=%22http://picasaweb.google.com/lh/photo/sA5TskBB9i_19FsG_H_4Ww?feat=embedwebsite%22%3E%3Ci mg%20src=%22http://lh5.ggpht.com/_AyIpx4vjLxg/TKJoWE7KGJI/AAAAAAAADtE/_5Ou-iIExCs/s800/2x%20mov%20VS%2010x%20LCD.JPG%22%20/%3E%3C/a%3E%3C/td%3E%3C/tr%3E%3Ctr%3E%3Ctd%20style=%22font-family:arial,sans-serif;%20font-size:11px;%20text-align:right%22%3EFrom%20%3Ca%20href=%22http://picasaweb.google.com/fraustosound/Science?feat=embedwebsite%22%3EScience%3C/a%3E%3C/td%3E%3C/tr%3E%3C/table%3E

09-28-2010, 04:04 PM
I believe it's because the line skipping or pixel bining method they use (not sure which) is activated during regular movie recording (not for stills). But when you do a zoom it's not. So...whatever method they are using.....sucks. :)

09-28-2010, 04:21 PM
Yes, this is well known. The stills resolution of the camera is vastly higher than its recorded video. The 5x and 10x are apparently windowing the sensor at full resolution (i.e., seeing a small portion of the sensor) but when you go to record video, it has to try to sample off the whole sensor, and the guess is that it skips lines and/or bins pixels in order to do that.

A simple experiment to show how much sharper things could theoretically be, would be to shoot a still, then resize that still down to 1920x1080 and compare it to what you get when you shoot video of the same subject.

09-28-2010, 04:34 PM
Stu Maschwitz has a demo of this on his blog:


09-28-2010, 05:09 PM
Canon could have done much better with this hardware had they wanted to.

09-28-2010, 05:47 PM
No they couldn't. That's the whole point. When you're in 5x or 10x, you're not seeing the whole sensor, you're only seeing a very small window of it.

The only time the whole sensor is used is in stills mode.

To scan the whole frame, at video speeds, they can't use all the sensor data -- it's too much to ask, so they skip or bin or whatever they need to do, to throw away 90% of what the sensor is capturing.

If you want a camcorder that scans the entire frame at video frame rates, then you're looking at a Red One -- and that costs 10x as much.

The Canon (and all other DSLRs) employ some cheats to get the massive amount of data down to a reasonable level. It's what they had to do, at the price point they had to hit.

09-28-2010, 05:59 PM
What everyone else said.

It brings to mind how awesome the VGA crop mode is on the 550D/t2i. I would love a 1080p crop mode of the sensor... although noisier, imagine no aliasing and imagine some deeper focus fun ;)

09-29-2010, 08:55 AM
Would it be possible to add a 1920x1080 "windowed mode" via firmware alone?