4000XCM에 관심을 가지고 야후 SBIG 그룹을 살펴보았더니 다음과 같은 글이 있어서 여기에 옮겨 봅니다.
아래 글의 저자(Craig)는 사람들이 흔히 원샷컬러CCD가 R1,G2, B1 총 4개의 픽셀로 나뉘어지므로 해상도가
전체 화소수의 1/4 밖에 안된다고 하는 말은 틀린 것이며, 가장 나쁘게 보아도 1/2 이상의 해상도는 가진다고 합니다.
그 이유는 Resampling algorithms에 따라 달라지기 때문이라는군요.
실제 현실에서 해상도의 손실이 있기는 있으나 매우 적다는 결론입니다.
샘플이미지도 함께 제시되어 있어 자세히 읽어 보시면 원샷컬러CCD에 대한 이해가 높아질 것 같습니다.
Resampling, resolution, one-shot -- was Re: NEWS ST-4000XCM vs ST-10XME
--- In SBIG@yahoogroups.com, George Hall <george2003@...> wrote:
>
> Stan,
>
> Concerning the 2 advantages of the one-shot color cameras, Isn't the
> resolution of the one shot color camera reduced to 1/4 of the mono
> camera with the same pixel size due to the sparse spacing of the pixels?
>
> George
George,
I'm not Stan, but I've worked a lot on image reconstruction from
one-shot CCDs and their Color Filter Arrays. The 1/4 resolution
notion is pure hype (or anti-hype). At worst, it should be 1/2. When
we speak of resolution, we speak of the ability to resolve small
features. Even if we consider each 2x2 area of the sensor as a single
pixel in color (using 1R, 2G, and 1B) we have half the horizontal and
half the vertical resolution. We don't have 1/4 the resolution.
(Imagine you imaged a set of fine vertical lines. You have "half" as
many pixels horizontally to resolve these so you have half the
resolution).
So 1/2 is the worst-case scenario. In practice, we're a lot better
than that. If all we did were to image random noise, we'd not do any
better than half. But, in truth, we image real objects. Real objects
have pixels that have structure when you compare them to nearby
pixels. This structure can be used to bring back resolution.
Consider the images shown here:
http://www.stark-labs.com/depot/images/resample2.png
In the center, we have the original. To make each of the others, a
2x2 bin was done and then various algorithms were used to turn it back
into the original size (up-sampling by 2x). Some are clearly better
than others. The "box" shows the 2x resolution loss. Many of the
others show the loss is not nearly so bad.
Resampling is at the heart of CFA reconstruction algorithms. I have
an article in Astrophoto Insight that goes over several of them
("Debayering Demystified"). More detail is present there than I can
cover here, of course, but here is a relevant figure:
http://www.stark-labs.com/craig/images/M4_comparison.png
We have the original and several attempts to reconstruct the original
after it has been passed through a Bayer matrix. So, I threw away all
but one color in each pixel in the original and re-constituted it
using several algorithms. While none are perfect, the VNG is quite
good and clearly superior to the Nearest Neighbor (which is a 2x
resolution loss).
From this, I say not only can we not support the conclusion that there
is only 1/4 as much resolution in one-shot as in mono, but that we
can't even say there is only 1/2 as much. For real-life data, there
is a loss, but it is quite small.
One-shot cameras do have their downsides, but the resolution loss is
nowhere near as severe as many believe.
Craig
(Disclaimer: I am the author of Nebulosity - referred to in some of
the images noted above).
아래 글의 저자(Craig)는 사람들이 흔히 원샷컬러CCD가 R1,G2, B1 총 4개의 픽셀로 나뉘어지므로 해상도가
전체 화소수의 1/4 밖에 안된다고 하는 말은 틀린 것이며, 가장 나쁘게 보아도 1/2 이상의 해상도는 가진다고 합니다.
그 이유는 Resampling algorithms에 따라 달라지기 때문이라는군요.
실제 현실에서 해상도의 손실이 있기는 있으나 매우 적다는 결론입니다.
샘플이미지도 함께 제시되어 있어 자세히 읽어 보시면 원샷컬러CCD에 대한 이해가 높아질 것 같습니다.
Resampling, resolution, one-shot -- was Re: NEWS ST-4000XCM vs ST-10XME
--- In SBIG@yahoogroups.com, George Hall <george2003@...> wrote:
>
> Stan,
>
> Concerning the 2 advantages of the one-shot color cameras, Isn't the
> resolution of the one shot color camera reduced to 1/4 of the mono
> camera with the same pixel size due to the sparse spacing of the pixels?
>
> George
George,
I'm not Stan, but I've worked a lot on image reconstruction from
one-shot CCDs and their Color Filter Arrays. The 1/4 resolution
notion is pure hype (or anti-hype). At worst, it should be 1/2. When
we speak of resolution, we speak of the ability to resolve small
features. Even if we consider each 2x2 area of the sensor as a single
pixel in color (using 1R, 2G, and 1B) we have half the horizontal and
half the vertical resolution. We don't have 1/4 the resolution.
(Imagine you imaged a set of fine vertical lines. You have "half" as
many pixels horizontally to resolve these so you have half the
resolution).
So 1/2 is the worst-case scenario. In practice, we're a lot better
than that. If all we did were to image random noise, we'd not do any
better than half. But, in truth, we image real objects. Real objects
have pixels that have structure when you compare them to nearby
pixels. This structure can be used to bring back resolution.
Consider the images shown here:
http://www.stark-labs.com/depot/images/resample2.png
In the center, we have the original. To make each of the others, a
2x2 bin was done and then various algorithms were used to turn it back
into the original size (up-sampling by 2x). Some are clearly better
than others. The "box" shows the 2x resolution loss. Many of the
others show the loss is not nearly so bad.
Resampling is at the heart of CFA reconstruction algorithms. I have
an article in Astrophoto Insight that goes over several of them
("Debayering Demystified"). More detail is present there than I can
cover here, of course, but here is a relevant figure:
http://www.stark-labs.com/craig/images/M4_comparison.png
We have the original and several attempts to reconstruct the original
after it has been passed through a Bayer matrix. So, I threw away all
but one color in each pixel in the original and re-constituted it
using several algorithms. While none are perfect, the VNG is quite
good and clearly superior to the Nearest Neighbor (which is a 2x
resolution loss).
From this, I say not only can we not support the conclusion that there
is only 1/4 as much resolution in one-shot as in mono, but that we
can't even say there is only 1/2 as much. For real-life data, there
is a loss, but it is quite small.
One-shot cameras do have their downsides, but the resolution loss is
nowhere near as severe as many believe.
Craig
(Disclaimer: I am the author of Nebulosity - referred to in some of
the images noted above).
