New blog about Full HD 'digital zoom' on DSLRs using crop - a viable option?

[Nino Leitner]

Hey guys,

I just wrote up a new blog post on the theory of a Full HD 'digital zoom' on DSLRs using crop, and I'd like to have a discussion on that topic - is it viable? Is my theory right?

Please check it out here, I also tried to explain how video acquisition with DSLRs works now and how such a crop could work:
Part 2: Full HD ?digital zoom? using crop ? a viable option?HD video with DSLRs – The good, the bad and the ugly. | Nino Film - Blog - Nino Leitner

Looking forward to your input on that matter!!

Nino

[Evan Donn]

The piece you're missing (and it's the problem with your suggestion in the first post that they make a 35mm 1920x1080 sensor) is that since these are single chip cameras - and will remain so due to the sensor size - you have to have more actual samples than your final resolution in order to get all three colors. You generally use a bayer filter on the sensor:

Bayer filter - Wikipedia, the free encyclopedia

I think at minimum you'll need twice as many photosites as your final resolution. In reality you'll need more if you want to maximize the final image resolution - you reduce aliasing with an optical low pass filter which also reduces resolution, so you want to start by sampling greater resolution than your final image.

So you could take just a 1920x1080 sample out of the center of the sensor, but you'll end up with an image that resolves significantly lower resolution than that. You could use a larger window but then you'd be back to line skipping. Plus, the bayer filter is an actual filter on top of the sensor - which means certain photosites can only be used for certain colors - so you can't necessarily chose an arbitrary segment of the sensor and have it work properly.

And there's the issue of depth of field - as you mentioned in the article it technically doesn't change. And in theory there's no difference between theory and practice, but in practice there is. In practice once you take a 3x crop you either have to move your camera 3x as far from the subject or use a much wider lens in order to get the same shot - you end up with the same situation as a small-chip camera where in order to get even a 'normal' field of view you have to use extremely wide angle lenses. That results in greater DOF and cost - and forget about getting true wide angle shots. The reality is once you do what you've suggested there would be little or no advantage over using a typical 3 chip 1/3" camera.

[Nino Leitner]

Thanks Evan, your input regarding the bayer filter sounds very compelling. I didn't think of that. Can you think of any way that this filter can be somehow planted into the camera and switched on and off as we need it? I guess that's a bit far-fetched, but maybe ...

But I think you misunderstood what I think the 1920x1080 crop in the middle of the sensor would be good for - I didn't think of it as an ALTERNATIVE to line skipping, but rather an ADDITION. I was mainly thinking about the current generations of cameras and how they could be enhanced using software only (firmware) to get more features. And I thought it would be nice if we could make use of all those pixels on the sensor and get a "fake" telephoto mode in addition to the standard line skipping mode - which of course cannot be replaced if we want to use the entire area of the sensor (which of course we want).

[Evan Donn]

Originally Posted by Nino Leitner

Thanks Evan, your input regarding the bayer filter sounds very compelling. I didn't think of that. Can you think of any way that this filter can be somehow planted into the camera and switched on and off as we need it? I guess that's a bit far-fetched, but maybe ...

Maybe I wasn't clear - this is how the camera works already. This is how single chip cameras - video or still - produce full color images, so the filter is already in the camera and there would be no reason to want to switch it on or off. It's also laid out for the full 21mp sensor, and this can have implications for the practicality of changing the line skipping count - red and blue are typically on adjacent lines on a sensor. Take every 3rd line and you'll still get alternating red/blue samples per line. Drop from 3 to two and you'll end up skipping one or the other color entirely.

But I think you misunderstood what I think the 1920x1080 crop in the middle of the sensor would be good for - I didn't think of it as an ALTERNATIVE to line skipping, but rather an ADDITION.

I get that, but I'm saying it wouldn't work very well, at least not as you've described, because of the bayer filter. You would likely get much better video out of any of canon's small, cheap HD cameras. It might be possible to do less of a crop - but to preserve sufficient photosites for the bayer pattern I don't think you could crop in much more than about 30%. Once you go beyond that you would be giving up resolution, plus there's the issue I mentioned with changing the line skipping frequency.

[Don Miller]

Originally Posted by Evan Donn

The piece you're missing (and it's the problem with your suggestion in the first post that they make a 35mm 1920x1080 sensor) is that since these are single chip cameras - and will remain so due to the sensor size - you have to have more actual samples than your final resolution in order to get all three colors. ...................

Not true, if I understand you correctly. The color of each pixel is estimated by the color values of surrounding pixels. Final image size matches bayer sensor size.

[Evan Donn]

I'm not talking about image size - I'm talking about resolution. Yes, you can take 1920x1080 photosites and produce a 1920x1080 image with a bayer sensor - but the actual detail resolved is going to be lower if half your image data is interpolated. Maybe it doesn't matter though - as it is now the camera doesn't achieve very high resolution (700 lines?). Considering the amount of data it has to work with (even sampling every third line) I would expect better - to me it indicates that the final resolution is largely a result of the processing of the image data rather than the sensor itself. But combine that subpar processing with a much smaller sample size (in a 1920x1080 windowed scenario) and I would expect the final result to be roughly equivalent to SD video upscaled to HD.