jackweng wrote:
DXO比你有說服力耶...(恕刪)
謝謝你的提示,這裡看過很多次了,沒什麼改變,後面的公式是這兩年被質疑才補上,以前是空的...
最後一次回應你，來說文解字吧....
--------------------------------------------------------
Consider, for example, the Canon EOS 350D, a model released in February 2005 and a very good camera for expert amateurs, and the Canon EOS 1Ds Mark III, a more recent (August 2007) professional DSLR camera, with a 21Mpix sensor. Both sensors have the same pixel pitch (6.4µm), and measurements of SNR on the RAW images under equivalent shooting conditions give similar results: 35.25dB for the 350D, 35.05dB for the 1Ds Mark III.
---------------------------------------------------------
這一段他再告訴你他們測試上不同片幅不同像素的機器在相同的像素大小下(6.4µm),有相近的躁訊比.
這部分螢幕上100%檢視的時候很容易觀察到..........
---------------------------------------------------------
Suppose now that the images have exactly the same field of view (by changing the lens focal length) and are printed on 30x20cm paper with the same 300dpi printer. Because of the very high resolution of the 1Ds Mark III, the printer will downsample the image and decrease the noise, giving a clear advantage of about 3dB to the 1Ds Mark III.
----------------------------------------------------------
這一段就是他的立論基礎認為以30x20cm 300dpi印出, 1Ds Mark III躁訊比有 3dB 增益.
1.沒有說明如何測試與測量印在紙上的躁訊比,這3dB哪裡來的.
2.沒有說明印出的設備與環境條件可提供驗證.
第二頁
----------------------------------------------------------
For a given exposure time, each smaller pixel receives four times less light than a large pixel—the equivalent of reducing exposure time by a factor of four. So to get the same sensor response, exposure time needs to be multiplied by four, which means that the ISO sensitivity of the high-resolution sensor is four times less than the low-resolution sensor.
Now assume that the same exposure times and identical ISO settings are used with a low-resolution camera and with a high-resolution camera having four times as many pixels. Since each high-resolution pixel is intrinsically less sensitive, a higher gain (either analog or digital) is applied to the signal, yielding more noise.
Let I denote the gray level on the sensor, and sL(I ) and sH(I ) the standard deviation on the low- and high-resolution sensors, respectively. With equivalent technology, the high-resolution sensor has more noise because of the higher gain: sL(I ) < sH(I ).
----------------------------------------------------------
這段再告訴你高像素必須有高的增益,可能是數位增益或是模擬的.......
----------------------------------------------------------
However, the four high-resolution neighboring pixels can be averaged out to form a low-resolution pixel. The statistical formula below shows the noise yield for the downsampled image:
The new SNR is
The loss of resolution produces a better SNR. We now have two images at the same resolution and shot in similar conditions. When printing with the same printer using the same format, it is more relevant to compare sL(I ) and sH(I ) + 6dB.
-----------------------------------------------------------
這邊又繼續告訴你高像素要+6dB的增益才會等同於低像素的躁訊比,但也沒告訴你這6dB哪裡來的.
第三頁
-----------------------------------------------------------
The argument above provides a method to compare cameras with different resolutions. A high-resolution camera can still be turned into a low-resolution camera by averaging its pixels.
In order to compare all cameras and not just pairs, a reference resolution can be chosen and equivalent SNRs at this resolution are computed. Let Nref denote this reference resolution. Consider a camera with N Mpix. To obtain an image with Nref pixels, N / Nref pixels need to be averaged out to produce a single pixel in the output image, so the noise decreases by a factor of . We then deduce that SNRref at resolution Nref can be obtained from the nominal SNR (in dB) using the formula
Notice that if the reference resolution Nref is very large, or if the initial sensor has a low resolution, the normalized SNR can actually be lower than the nominal SNR.
------------------------------------------------------------
其中這段
------------------------------------------------------------
Consider a camera with N Mpix. To obtain an image with Nref pixels, N / Nref pixels need to be averaged out to produce a single pixel in the output image, so the noise decreases by a factor of . We then deduce that SNRref at resolution Nref can be obtained from the nominal SNR (in dB) using the formula
Notice that if the reference resolution Nref is very large, or if the initial sensor has a low resolution, the normalized SNR can actually be lower than the nominal SNR.
------------------------------------------------------------
這段他告訴你投放到Nref pixels圖像上(印出或是縮圖),很多像素會平均到一個像素上,所以認為會減低噪音.所以他用原始的SNR去做加成計算.
1.自己都說了像素會平均,只降低噪音沒降低訊號??這也太神奇??
2.同樣無法驗證.
再下來就不翻了..............
