What does 640x480 mean?
Martin B. Brilliant
The Agfa ePhoto Smile and the Largan Lmini 350 were both under $100
and promised 640x480 resolution. I never heard of Largan, and Agfa has
a reputation dating back more than 100 years, so I bought the Agfa.
I still don't have the camera I need. Based on shots of a
high-contrast test target, the real overall image resolution is about
160x120 (it took 8 pixels in the 640x480 output image to resolve a
line pair).
Agfa advertises "640x480 resolution with PhotoGenie"; the Largan
camera was advertised as "true 640x480 resolution." What does all this
mean? Agfa lied; is Largan lying?
Marty
Martin B. Brilliant at home in Holmdel, NJ
http://www.netlabs.net/hp/marty
Marvin Margoshes
"Martin B. Brilliant" <ma...@technologist.com> wrote in message
news:39c8eff4...@news.netlabs.net...
By convention, 640X480 reflects the number of pixels on the X and Y axes.
Translating that into resolution on a test target is complex. It can be
affected by the quality of the lens, the focus, the software, etc. Mnay
cameras compress the data file to save storage space, which results in some
loss of resolution. Another factor may be the resolution of the image
display - monitor or printer.
It takes two pixels to make a resolution unit, so the best you can get with
640X480 pixels is a resolution of 320X240 lines or 160X120 line pairs -
which is what you observed.
DGiunti
B. Brilliant) writes:
>I need a low-cost digital camera to take pictures to post on the WWW.
>The Agfa ePhoto Smile and the Largan Lmini 350 were both under $100
>and promised 640x480 resolution. I never heard of Largan, and Agfa has
>a reputation dating back more than 100 years, so I bought the Agfa.
>
>I still don't have the camera I need. Based on shots of a
>high-contrast test target, the real overall image resolution is about
>160x120 (it took 8 pixels in the 640x480 output image to resolve a
>line pair).
>
>Agfa advertises "640x480 resolution with PhotoGenie"; the Largan
>camera was advertised as "true 640x480 resolution." What does all this
>mean? Agfa lied; is Largan lying?
Many camera makers use this type of hyped resolution for their wares. Agfa
is an old hand at this. The camera you have interpolates 320x240 to 640x480 in
software. I have an Agfa 1280 myself. It's ccd is 1024x768 and photogene
grabs it's highest res and converts it to 1280x960. I don't use this feature
much. Otherwise my Agfa is a really nice camera for web use.
The Largan may have a 640x480 ccd. I don't think that either of them has a
flash though and that is a handy thing to have for web sharing. I think that
you may find a review of the Largan at http://shopper.com/ I recall seeing
one while I was shopping for my first digital.
You may want to do more shopping and come back with a flash. I think that
low end for 640x480 is about $150 though.
good luck in you search
David Giunti email: DGi...@aol.community
What is the question? Gertrude Stein's last words
No one mouth is big enough to utter the whole thing. Alan Watts
On Display in the UK http://www.web-gallery.co.uk
Mark Grebner
> It takes two pixels to make a resolution unit, so the best you can get
> with 640X480 pixels is a resolution of 320X240 lines or 160X120 line
That's exactly right. It's an interesting problem with typical digital
cameras that they are perversely poor at photographing line pairs,
because the data isn't dealt with one pixel at a time, but only after
the data from 2x2 groups of pixels has been combined. Since every group
of 2x2 pixels will see half black and half white when the number of
lines is equal to the pixel count (that is 640 pixels, 640 lines, 320
line pairs) the camera doesn't find any image at all under such
conditions.
It's not completely fair to blame camera-manufacturer-hype for the
problem. Targets of equal-size line pairs turn out to be a very
difficult test for most digital cameras. They do much better at
resolving other patterns that would logically seem to be equivalent.
In particular, if you try photographing 214 vertical line pairs (so the
line count is equal to 2/3 of the number of pixels) you'll discover that
the camera captures a slightly muddy image of 214 line pairs. At higher
densities, there is still some image, but aliasing occurs, and the count
of line pairs is distorted.
The problem is NOT the lens or the JPEG compression routines. They
cause problems of their own, but this is a fundamental problem having to
do with the way the (typical) digital camera manipulates pixel-level
data.
Sent via Deja.com http://www.deja.com/
Before you buy.
Martin B. Brilliant
at 640x480, 4 pixels for a line (= picture element), for 160x120
picture elements total. Marvin Margoshes said he expects 160x120 line
pairs, which is 320x240 picture elements.
So I'm getting only half the resolution I should get, even after you
figure that it takes two pixels to represent a line, which is 4 pixels
per line pair. I get (as I said) 8 pixels per line pair.
On Wed, 20 Sep 2000 14:18:05 -0400, "Marvin Margoshes"
<physche...@telocity.com> wrote:
>
> "Martin B. Brilliant" <ma...@technologist.com> wrote in message
> news:39c8eff4...@news.netlabs.net...
> > I need a low-cost digital camera to take pictures to post on the WWW.
> > The Agfa ePhoto Smile and the Largan Lmini 350 were both under $100
> > and promised 640x480 resolution. I never heard of Largan, and Agfa has
> > a reputation dating back more than 100 years, so I bought the Agfa.
> >
> > I still don't have the camera I need. Based on shots of a
> > high-contrast test target, the real overall image resolution is about
> > 160x120 (it took 8 pixels in the 640x480 output image to resolve a
> > line pair).
> >
> > Agfa advertises "640x480 resolution with PhotoGenie"; the Largan
> > camera was advertised as "true 640x480 resolution." What does all this
> > mean? Agfa lied; is Largan lying?
> >
> > Marty
> > Martin B. Brilliant at home in Holmdel, NJ
> > http://www.netlabs.net/hp/marty
>
> By convention, 640X480 reflects the number of pixels on the X and Y axes.
> Translating that into resolution on a test target is complex. It can be
> affected by the quality of the lens, the focus, the software, etc. Mnay
> cameras compress the data file to save storage space, which results in some
> loss of resolution. Another factor may be the resolution of the image
> display - monitor or printer.
>
> It takes two pixels to make a resolution unit, so the best you can get with
> 640X480 pixels is a resolution of 320X240 lines or 160X120 line pairs -
> which is what you observed.
>
>
Martin B. Brilliant
640x480 camera has only a 130,000 pixel CCD, but their PhotoGenie
makes up for it. Makes it up out of thin air.
TigerDirect (www.tigerdirect.com) has both the Largan Lmini 350 and
the Agfa ePhoto Smile for well under $100. And both have flash.
I did find a camera on the web that claimed optical resolution of
640x480, but at a price over $250. Not what I wanted.
> ... I have an Agfa 1280 myself. It's ccd is 1024x768 and photogene
> grabs it's highest res and converts it to 1280x960. I don't use this feature
That's a 56 percent increase courtesy of PhotoGenie, and it's tame
compared to what PhotoGenie does for the ePhoto Smile -- 640x480 out
of a 130,000 pixel CCD, for a 136 percent increase. And I can't opt
out.
On 20 Sep 2000 18:43:53 GMT, dgi...@aol.community (DGiunti) wrote:
> In article <39c8eff4...@news.netlabs.net>, ma...@technologist.com (Martin
> B. Brilliant) writes:
>
> >I need a low-cost digital camera to take pictures to post on the WWW.
> >The Agfa ePhoto Smile and the Largan Lmini 350 were both under $100
> >and promised 640x480 resolution. I never heard of Largan, and Agfa has
> >a reputation dating back more than 100 years, so I bought the Agfa.
> >
> >I still don't have the camera I need. Based on shots of a
> >high-contrast test target, the real overall image resolution is about
> >160x120 (it took 8 pixels in the 640x480 output image to resolve a
> >line pair).
> >
> >Agfa advertises "640x480 resolution with PhotoGenie"; the Largan
> >camera was advertised as "true 640x480 resolution." What does all this
> >mean? Agfa lied; is Largan lying?
>
> Many camera makers use this type of hyped resolution for their wares. Agfa
> is an old hand at this. The camera you have interpolates 320x240 to 640x480 in
> software. I have an Agfa 1280 myself. It's ccd is 1024x768 and photogene
> grabs it's highest res and converts it to 1280x960. I don't use this feature
> much. Otherwise my Agfa is a really nice camera for web use.
>
> The Largan may have a 640x480 ccd. I don't think that either of them has a
> flash though and that is a handy thing to have for web sharing. I think that
> you may find a review of the Largan at http://shopper.com/ I recall seeing
> one while I was shopping for my first digital.
>
> You may want to do more shopping and come back with a flash. I think that
> low end for 640x480 is about $150 though.
>
> good luck in you search
>
> David Giunti email: DGi...@aol.community
> What is the question? Gertrude Stein's last words
> No one mouth is big enough to utter the whole thing. Alan Watts
>
> On Display in the UK http://www.web-gallery.co.uk
Martin B. Brilliant
640x480 picture elements, but only 320x240 picture elements, because
it takes a 2x2 group to get all the information for one picture
element. Therefore, building a camera with a 640x480 CCD and saying it
gives 640x480 pixel resolution is hype.
This all strikes back at the comparison of digital cameras with film.
Take a fair to middling 35mm camera with only 30 line-pairs/mm
resolution across 36mm x 24mm. That's 60x36 by 60x24 actual pixels, or
2160x1440 end-to-end picture resolution. If a digital camera is to
match that, using 2x2 CCD elements per pixel, it needs a 4320x2880
CCD. Who's marketing that, and at what price?
On Wed, 20 Sep 2000 23:31:01 GMT, Mark Grebner
<mark_g...@my-deja.com> wrote:
>
> > It takes two pixels to make a resolution unit, so the best you can get
> > with 640X480 pixels is a resolution of 320X240 lines or 160X120 line
Mark Grebner
> So basically what I'm hearing is that a 640x480 CCD will not resolve
> 640x480 picture elements, but only 320x240 picture elements, because
> it takes a 2x2 group to get all the information for one picture
> element. Therefore, building a camera with a 640x480 CCD and saying it
> gives 640x480 pixel resolution is hype.
all that's available is a sliding average. To simplify to a single
dimension, the first apparent pixel is derived from the pair of points
in columns 1 & 2. The second apparent pixel comes from 2 & 3, then 3 &
4 and so on.
If you present 320 line pairs to that array of sensors, maybe column 1
will be all black, but if it is, column 2 is all white. So the
calculated value for the first apparent pixel is 50% gray. And so are
all the other pixels.
Equal-size alternating bands are just really a difficult problem for
that arrangement. Imagine a different kind of target: a black bar on a
white background, whose length you want to measure. For that kind of
use, less damages is done by averaging the sensor data into 2x2 packets.
In fact, your ability to perceive the length would only be slightly
worse than if you had access to the original 1x1 data.
I read on this forum a claim that this 2x2 method results in a
resolution about 70% as good as what I'm calling 1x1 would be.
By the way, one reason for the 2x2 blending is that each sensor pixel is
covered by red, blue, or green filtering, which are used to create color
values for each pixel. Each 2x2 group includes 2 green pixels (situated
diagonally), one red, and one blue. I suppose if the camera were
monochrome, it would be fairly easy to preserve 1x1 data - but it would
be hard to do so while also gathering color data.
For my work - which is copying large volumes of black-on-white documents
located in public buildings - I really wish there were a 1x1 mode
available on some camera somewhere. But I've learned to live with the
reality that typical users, who drive the market and make it possible to
sell cameras cheaply, need color.
I've photographed more than my share of line-pair targets using various
cameras, and at first I was as angry as the original poster here about
the over-hyped resolution claims. But I've mellowed over time, as I've
learned to understand and live with the limitations built into available
equipment.
Marvin Margoshes
> So basically what I'm hearing is that a 640x480 CCD will not resolve
> 640x480 picture elements, but only 320x240 picture elements, because
> it takes a 2x2 group to get all the information for one picture
> element. Therefore, building a camera with a 640x480 CCD and saying it
> gives 640x480 pixel resolution is hype.
>
it is 640 pixels long, that is the number of picture elements. From
sampling theory (usually applied to Fourier transform methods but not
limited to that) two picture elements to make one resolution element. In
photography, it is common to define resolution into line pairs - not single
lines - and that requires not 2 but 4 picture elements.
> This all strikes back at the comparison of digital cameras with film.
> Take a fair to middling 35mm camera with only 30 line-pairs/mm
> resolution across 36mm x 24mm. That's 60x36 by 60x24 actual pixels, or
> 2160x1440 end-to-end picture resolution. If a digital camera is to
> match that, using 2x2 CCD elements per pixel, it needs a 4320x2880
> CCD. Who's marketing that, and at what price?
>
There is a word "specsmanship" - the distortion of specifications for
marketing purposes.
If you have a budget like the military or NASA, you can put a digital camera
in a satellite that will give incredibly high-resolution pictures of the
earths surface or distant galaxies.
Marvin Margoshes
> Somebody misread me. I said it took 8 pixels to represent a line pair
> at 640x480, 4 pixels for a line (= picture element), for 160x120
> picture elements total. Marvin Margoshes said he expects 160x120 line
> pairs, which is 320x240 picture elements.
>
> So I'm getting only half the resolution I should get, even after you
> figure that it takes two pixels to represent a line, which is 4 pixels
> per line pair. I get (as I said) 8 pixels per line pair.
>
640/160 = 4, not 8.
John
If you buy a digital camera for $640, it will drop down to $480 30 days
later.
John
DGiunti
B. Brilliant) writes:
>> ... I have an Agfa 1280 myself. It's ccd is 1024x768 and photogene
>> grabs it's highest res and converts it to 1280x960. I don't use this
>feature
>
>That's a 56 percent increase courtesy of PhotoGenie, and it's tame
>compared to what PhotoGenie does for the ePhoto Smile -- 640x480 out
>of a 130,000 pixel CCD, for a 136 percent increase. And I can't opt
>out.
You may be able to 'opt out' of this. Agfa cameras are among those hacked
with PD freeware. I use Came from
http://www.butaman.ne.jp:8000/~tsuruzoh/
to download the highest res images that PhotoGenie would inflate and just work
with the extra data in it's 1024x768 images. You may be able to do this with
the smile, and surely if it has a 320x240 sized mode. Came may get your images
in their more natural size, but it's still kind of a small pic.
Martin B. Brilliant
I can't find where I said that. In my message of 9/20/00 titled "What
does 640x480 mean," second paragraph, I wrote
> ... the real overall image resolution is about
> 160x120 (it took 8 pixels in the 640x480 output image to resolve a
> line pair).
In case it's not perfectly clear that 160x120 means pixels, not line
pairs, look at the remark in parentheses, divide 640 (pixels) by 8
(pixels per line pair) and the result (algebraically computing the
dimensionality, if you will) is 80 line pairs, which means 160 pixels,
not 160 line pairs.
This reminds me of the angle of a pie, in radians. One pie is 2pi, so
half a pie is pi, pi/2 is a quarter of a pie, etc.
The bottom line: a 640x480 CCD, smudged by 2x2 for color-averaging,
should give you 320x240 pixels, or 160x120 line pairs. I got 80x60
line pairs (extrapolating from the resolution at the center, even
though the image obviously deteriorates toward the corners). That's
bad.
Also, as pointed out in another article in this thread, the 2x2
averaging does not actually reduce the resolution by half in both
directions, because the averaging is a sliding average. There are
still 639x479 data points after averaging. Theoretically, the
averaging does not destroy a lot of information, because given these
639x479 data points and the pixel values along two adjacent edges (as
"boundary conditions") you could reconstruct the original 640x480
pixels. What does destroy information in the Agfa ePhoto Smile is the
lousy optical resolution.
Martin B. Brilliant
<physche...@telocity.com> wrote:
>
> "Martin B. Brilliant" <ma...@technologist.com> wrote in message
> news:39c966a9...@news.netlabs.net...
> > So basically what I'm hearing is that a 640x480 CCD will not resolve
> > 640x480 picture elements, but only 320x240 picture elements, because
> > it takes a 2x2 group to get all the information for one picture
> > element. Therefore, building a camera with a 640x480 CCD and saying it
> > gives 640x480 pixel resolution is hype.
> >
> To make things easier, let's concentrate on one of the two dimensions. If
> it is 640 pixels long, that is the number of picture elements. From
> sampling theory (usually applied to Fourier transform methods but not
> limited to that) two picture elements to make one resolution element. In
> photography, it is common to define resolution into line pairs - not single
> lines - and that requires not 2 but 4 picture elements.
Tell me about sampling theory - I still remember some. The sampling
rate must be at least twice the highest frequency to be recorded.
Let's just say the horizontal dimension of the raster, 640 pixels, is
one centimeter. The sampling frequency in the horizontal direction is
therefore 640/cm. This can record a sinusoid of 320 cycles/cm, which
will resolve 320 vertical line-pairs/cm. Horizontal scanning finds two
picture elements (one black, one white) for each line pair, or 640
picture elements across the horizontal dimension.
It still takes a 2x2 group of pixels (a.k.a. picture elements) in the
CCD to get one picture element (a.k.a. pixel) in the output picture.
But other messages in this thread have pointed out that the 2x2 groups
are overlapping, so that the output picture still has, not 320x240
pixels, but nearly 640x480 pixels. The trouble is, these 640x480
pixels are not independent, so they can't properly be called
"elements." Theoretically, however, the original resolution can be
nearly recovered by sharpening the output picture.
Martin B. Brilliant
<mark_g...@my-deja.com> wrote:
> Well, only sort of. THere are actually 640x480 picture elements, but
> all that's available is a sliding average. To simplify to a single
> dimension, the first apparent pixel is derived from the pair of points
> in columns 1 & 2. The second apparent pixel comes from 2 & 3, then 3 &
> 4 and so on.
>
> If you present 320 line pairs to that array of sensors, maybe column 1
> will be all black, but if it is, column 2 is all white. So the
> calculated value for the first apparent pixel is 50% gray. And so are
> all the other pixels.
If the CCD has 640x480 elements, then there are 639x479 2x2 groups in
the sliding average. If you know enough boundary conditions -- for
instance, the pixel values along the original first column and first
row -- you can recompute all the original pixel values.
As above, say the first column is all black, and the second column is
all white, and suppose you know the edge values and all the 2x2
averages. Starting in the upper left corner, you know the first 2x2
average is medium gray. But you also know that two of the original
pixels in it are black, and one is white, so the one you weren't given
(2nd row, 2nd column) must be white. You can work along all the rows
and columns that way and reconstruct (sharpen) the entire image.
Unfortunately, the camera output doesn't include those 1119 boundary
values.
> ...
> For my work - which is copying large volumes of black-on-white documents
> located in public buildings - I really wish there were a 1x1 mode
> available on some camera somewhere. But I've learned to live with the
> reality that typical users, who drive the market and make it possible to
> sell cameras cheaply, need color.
If you know your documents have white space, then you may have enough
boundary conditions to reconstruct the original full-resolution
monochrome image. Theoretically, the output pixels can be white,
black, or three shades of gray, depending on how many black pixels
there are in the 2x2 group. If an output pixel is white, you know it's
an average of four white CCD pixels. If you have a light gray pixel
nearly surrounded by white pixels, you know where one original black
pixel is. It's a little like playing minesweeper - and just as
sensitive to uncertainties, due in this case to the paper not being
pure white, the ink not being pure black, and the CCD's having
different color sensitivities. But some software package ought to have
a sharpening algorithm, or a way to let you write your own, that will
get you a more readable output than the image you get straight from
the camera.
In my case, it's hopeless, because the output resolution of the Agfa
ePhoto Smile is limited by the optical system, not the 2x2 averaging.
The accompanying software applies a sharpening algorithm that makes
the output look ugly, but not readable -- and there's no way to turn
it off.
Martin B. Brilliant
camera." I guess CAME doesn't support the meager brain behind the
Smile.
DGiunti
B. Brilliant) writes:
>
>Thanks, but it didn't do anything for me. Kept telling me to "connect
>camera." I guess CAME doesn't support the meager brain behind the
>Smile.
You have to have the camera turned on with the serial cable connected for
came to "get" the camera info. I am not really familiar with all the error
conditions from Came since it worked right out of the box for me. But there
could be problems with the way that the serial ports are connected on the
machine.
From your conversation regarding resolution I don't think that little Smile
will be enough camera for you. You really have to throw lots of ccd pixels at
that problem to solve it. Unfortunately there are few real bargains out there
in the high res area. In fact the digital camera market seems to have
plateaued on the price front. The low end cameras have actually gone up in
price since I shopped for mine, and this is exactly the opposite from what has
been going on in the past couple of years.
Good Luck!
Dave
>On 22 Sep 2000 00:07:27 GMT, dgi...@aol.community (DGiunti) wrote:
>
>> In article <39c96385...@news.netlabs.net>, ma...@technologist.com
>(Martin
>> B. Brilliant) writes:
>>
>>
>> >> ... I have an Agfa 1280 myself. It's ccd is 1024x768 and photogene
>> >> grabs it's highest res and converts it to 1280x960. I don't use this
>> >feature
>> >
>> >That's a 56 percent increase courtesy of PhotoGenie, and it's tame
>> >compared to what PhotoGenie does for the ePhoto Smile -- 640x480 out
>> >of a 130,000 pixel CCD, for a 136 percent increase. And I can't opt
>> >out.
>>
>> You may be able to 'opt out' of this. Agfa cameras are among those
>hacked
>> with PD freeware. I use Came from
>>
>> http://www.butaman.ne.jp:8000/~tsuruzoh/
>
David Giunti email: DGi...@aol.community
Mark Grebner
> As above, say the first column is all black, and the second column is
> all white, and suppose you know the edge values and all the 2x2
> averages. Starting in the upper left corner, you know the first 2x2
> average is medium gray. But you also know that two of the original
> pixels in it are black, and one is white, so the one you weren't given
> (2nd row, 2nd column) must be white. You can work along all the rows
> and columns that way and reconstruct (sharpen) the entire image.
> Unfortunately, the camera output doesn't include those 1119 boundary
> values.
a few pixels away. That is, we're not talking about a formal logic
problem, but we're working with an actual physical system, where (for
example) the pixels don't mesh precisely with the target image, and the
misalignment changes over the image in many different ways.
Sharpening (IMO) always involves the risk of "creating" information.
Information which may be very clear, but also very wrong.
Martin B. Brilliant
> In article <39cf6350...@news.netlabs.net>, ma...@technologist.com (Martin
> B. Brilliant) writes:
>
> >
> >Thanks, but it didn't do anything for me. Kept telling me to "connect
> >camera." I guess CAME doesn't support the meager brain behind the
> >Smile.
>
> You have to have the camera turned on with the serial cable connected for
> came to "get" the camera info....
I know all about turning on the camera. I had a lot of "fun" making
the Agfa PhotoWise software (that came with the camera) work on my
Win95 system. It kept saying "cannot initialize camera connection" and
quitting before it finished loading. I finally set up a second (new,
fresh, and clean) installation of Win95 (booting from another primary
partition) on the same hardware. It wasn't the physical camera
connection at all, but (maybe) something in the system setup that
prevented the Agfa software from recognizing the serial port the
camera was connected to. Agfa tech support couldn't (or didn't want
to) figure it out.
> ... The low end cameras have actually gone up in
> price since I shopped for mine, and this is exactly the opposite from what has
> been going on in the past couple of years.
Not exactly. TigerDirect has two low-end cameras. The Agfa ePhoto
Smile (the one I bought) is now a reduced-price special, while the
Largan Lmini 350 (the one that was advertised as "true 640x480") is up
in price. They were the same price a few months ago. (I also pick the
wrong stocks.)
Martin B. Brilliant
<mark_g...@my-deja.com> wrote:
>
> > It takes two pixels to make a resolution unit, so the best you can get
> > with 640X480 pixels is a resolution of 320X240 lines or 160X120 line
> > pairs which is what you observed.
Actually, I observed (as I said originally) 160x120 pixels, or 80x60
line pairs.
> That's exactly right. It's an interesting problem with typical digital
> cameras that they are perversely poor at photographing line pairs,
> because the data isn't dealt with one pixel at a time, but only after
> the data from 2x2 groups of pixels has been combined. Since every group
> of 2x2 pixels will see half black and half white when the number of
> lines is equal to the pixel count (that is 640 pixels, 640 lines, 320
> line pairs) the camera doesn't find any image at all under such
> conditions.
It just dawned on me (after I made up a color resolution test chart)
that the difficulty of resolving line pairs in a digital camera is
totally irrelevant to the performance of the Agfa ePhoto Smile.
Theoretically, a 640x480 CCD without averaging ought to be able to
resolve half that number, or 320 x 240 line pairs. With averaging, it
can't do that, but it should resolve one-third that number, or about
213 x 160 line pairs. The Agfa ePhoto Smile resolves 80 x 60 line
pairs, which is about 1/3 of what the camera would resolve with
perfect optics.
Clearly it's the optical performance, not the way a CCD is used in a
digital camera, that's responsible for what I'm seeing (or not
seeing).