This explanation of DSLR video’s aliasing doesn’t seem to hold up.

The HD-sized images from these cameras are being downsized from a vastly larger frame; any aliasing due to the chip and its OLPFs would occur at the native resolution of the chip.  Resizing it down to HD resolution would vastly reduce (if not totally obliterate) this extremely fine aliasing.  In contradiction to his conclusions, Barry Green says the same thing in the linked article:

“...since web video is frequently scaled down in size (and scaling down helps to reduce aliasing artifacts) they released the 5D Mk II as-is…”

Scaling an image larger than 5000 x 3000 down to 1920 x 1080 should not result in more aliasing than a full-sized still.  And 1280 x 720 should look smoother still.  But in fact we don’t see that at all; most people think the 720 footage looks MORE aliased (although that’s probably because they’re blowing it up to the same size as the 1080 for viewing).

The aliasing in DSLR video almost certainly comes from one thing: a fast and crappy resizing algorithm.  What happens if you resize an image with no resampling?  Try it in your favorite image-editing program.  The result: wretched aliasing.

To achieve the necessary processing throughput to deliver “HD” video, these cameras are taking shortcuts.  They’re not downsampling the images well, and the result is aliasing galore.  They may even be doing some abbreviated deBayering to speed things up.  It almost certainly has nothing to do with optical low-pass filters, guys.

Posted by .(JavaScript must be enabled to view this email address)  on  12/09  at  03:13 AM

---

MobiusStrip and the aliasing article are both half right.

The article is correct that the problem is the lack of a low pass filter.  But it fails to note that the low pass filter could potentially be applied to the digital data (software) and need not be applied optically at the sensor level.  At least that is a possibility if the camera uses all of the sensor data and then resamples a full frame to HD resolution.  So it is quite possible - at least from an information theory standpoint - to have the DSLR highly optimized for still and video.  The question is how fast the sensor can offload data and how fast the camera can run it through a low pass (blur) digital filter.

MobiusStrip is quite possibly also correct in attributing the issue to how the image is resampled.  This is correct if the camera actually uses a full sensor frame of information and then resamples it. But for all we know, the camera may be taking every other line or every third line of image detail from the sensor and tossing out the in-between lines.  If that’s the case, then better resampling or software based low pass filtering will not be helpful. In this case the aliasing is introduced at capture and once introduced, aliasing cannot be removed without removing HUGE amounts of image detail.

The author of the article mentioned slightly soft focus to combat aliasing.  Two other things that might be worth trying are to use lenses that aren’t inherently sharp to begin with - like some third party generic brand glass - or to use very small apertures that introduce softening due to diffraction.  The other end of that scale is to use a fast lens wide open which tends to yield softening due to lens aberrations.  Most still camera lenses are significantly compromised for sharpness (there are eceptions) when shot wide open.

Posted by Jay Turberville  on  12/10  at  05:46 PM

---

Even with a “fast and crappy resizing algorithm” these cameras overheat. If they resized correctly they would turn down three seconds after you pushed record. These cameras are not intended for video. They have no decent ventilation. The cheap T2i even has a plastic body, which does not help. It is a known fact that the 7D with its metal body does not lock up as quickly, because the body dissipates heat better. For video these cameras need ventilation holes and maybe even a fan, like Panasonic did on its 60p consumer models. This, and less power-hungry chips, and then Canon et all will be able to improve filtering, scaling and encoding.

Posted by .(JavaScript must be enabled to view this email address)  on  08/27  at  06:21 PM

---

Wrote something about how the Canon do their line skipping a while back as well - check it out:
http://ninofilm.net/blog/2010/04/26/full_hd_crop_zoom/

Posted by Nino  on  08/28  at  12:44 PM

---

Name:

Email:

Location:

URL:

Smileys

Remember my personal information

Notify me of follow-up comments?

Submit the word you see below: