| |
In lugnet.space, Brian H. Nielsen wrote:
> In lugnet.space, Todd Lehman writes:
> > In lugnet.space, David Laswell writes:
> > > [...] Basically, digital zoom doesn't do anything that you can't
> > > accomplish by enlarging the image with MS Paint.
> >
> > I don't think that's quite correct. If your CCD or CMOS chip has greater
> > than 8 bits of depth on each spectral band (most do), then an on-camera
> > digital zoom should contain more information than a post-processed
> > zoom.
> >
> > For example, suppose you have a 10-bit CCD (common), 10x digital zoom,
> > and 5x optical zoom. Now if you take a 24-bit color JPEG photo at 10x
> > (digital zoom), download it from the camera, load it into Photoshop, convert
> > it to a 16-bit image, and resize the image to 50% (half size), that _should_
> > contain more detail than if you had just taken the photo at 5x (optical zoom).
> > If they look identical, then the camera has a really poor digital zoom.
>
> I think you are confusing color depth with image area resolution.
I don't think I am. :-) I'm not saying that the end result of a 2x-downsampled
10x digital zoom will look any better in 24-bit color than a native 5x optical
zoom would look in 24-bit color. It won't. What I'm saying is that a
2x-downsampled 10x digital zoom will contain _more detail_ and _more accuracy_
in *48-bit* color (i.e., high-quality color) than native 5x optical zoom would
look if converted from 24-bit to 48-bit color.
> If the CCD chip has an area of X-by-Y pixels and you are at maximum optical
> zoom, then any form of digital zoom requires interpolating between adjacent
> real pixels.
Right.
> The color depth of those real pixels does not help the interpolation process
> for the spatial data that lies between those real pixels.
Ah, but it does help -- especially in high-contrast areas. Think about this
in a more extreme case. Suppose you have an 8-bit (per color) CCD, and your
camera's output is 4 bits per color (so 12 bits total per pixel or 4096 colors)
instead of 8 bits per color (or 24 bits total per pixel or ~16 million colors).
And suppose your camera's maximum optical zoom is 5x and its maximum digital
zoom is 20x.
Now, when you take a 5x optical zoom photo and crop out the center 100x100
pixels, you've got an image with 100x100x12 bits (or about 15 KB) of information
content In contrast, when you take a 20x digital zoom photo and crop out the
center 400x400 pixels for 4x downsampling, you start with an image having
400x400x12 bits (~234 KB) of information content and you end up after
downsampling with an image having 100x100x20 bits (~24 KB) of information
content (provided of course that you downsample using high-precision
arithmetic). Did those extra bits of information come from thin air? Nope,
they came from the fact that the digital zoom did the interpolation using
8-bit arithmetic instead of 4-bit arithmetic.
> For example, if you're interpolating between the two numbers 11.111111
> and 33.333333 you've got a specifc linear gradient between them (22.222222).
> Adding extra digits of precision (11.11111111 and 33.33333333) does not
> change the gross value of the gradient, it only gives greater precision to
> the gradient (22.22222222).
A gradient computed with higher precision arithmetic contains less total error
than a gradient computed with lower precision. Doing Floyd-Steinberg on the
gradient helps a lot, but digital zooms don't to FS, so that's a moot point.
> Whether that interpolation takes place in the camera or on the PC doesn't
> help get more spatial detail between the real pixels in the zoomed area.
It's not a camera-vs-PC thing, it's a bit-depth thing. If you can get the
raw 10- or 12-bit CCD data from the camera (some cameras allow this), or a
48-bit JPEG from the camera instead of a 24-bit JPEG, then fine...there's no
advantage in using digital zoom. BUT if you habitually download 24-bit JPEG
images, then a digital zoom simply contains more totaly information content
_in the target area_ because less total information is lost in the conversion
from 10/12-bit down to 8-bit.
> In fact, if you do it on the camera you have to throw away all the real
> pixels outside the area you digitally zoom to, thus losing the opportunity
> to ever do so in the future.
This is for cases when optical zoom maxes out early and you'd be cropping
anyway. If your optical zoom maxes out at 3x and your subject in the file
you download is only 200x200 pixels, then I recommend that you crank up
the digital zoom to full, and get the subject in the frame as much as you can,
hopefully 400x400 or 600x600, and then downsample *that* to 200x200 pixels in
a 48-bit color space (16 bits per color). That's if you *really* want to
eek out every last possible bit of information content.
Digital zoom does have its uses.
--Todd
|
|
Message has 1 Reply:
 | | Re: taking good photos
|
| (...) While there is more color fideltiy at higher bit-depths, above a certain point the human eye can't tell the difference. 24-bit is also termed "true color" because it's 16.7 million color range is close to the limit of what the human eye can (...) (21 years ago, 2-Jun-03, to lugnet.space, lugnet.publish.photography)
|
Message is in Reply To:
| | Re: taking good photos
|
| (...) I think you are confusing color depth with image area resolution. If the CCD chip has an area of X-by-Y pixels and you are at maximum optical zoom, then any form of digital zoom requires interpolating between adjacent real pixels. The color (...) (22 years ago, 22-May-03, to lugnet.space, lugnet.publish.photography)
|
22 Messages in This Thread:
 
 
- Entire Thread on One Page:
- Nested:
All | Brief | Compact | Dots
Linear:
All | Brief | Compact
This Message and its Replies on One Page:
- Nested:
All | Brief | Compact | Dots
Linear:
All | Brief | Compact
|
|
|
|
