|
Forum: Troubleshooting and Beginner Help
09-18-2015, 05:38 AM
|
| |
Yesterday I did a test using the Pentax K-01 to see the actual levels in the wells.
These levels are stored the camera raw file, and they can be read using utilities.
In bright sunlight I set 100 iso, and used the -M 100mm macro lens to photograph first a gray card ( LAB(50,0,0) )
and then a bright red t shirt fabric. I used the macros lens to completely fill the sensor frame with similar values.
Both photos were near sunny 16, 125th and f/11 and the camerea was set to raw+jpg. The camera jpgs were well exposed and the histograms were between 2/4 and 3/4
I then pointed the camera towrd the sun and took a way overexposed photo. That was to saturate the wells for a check of the 100% values.
The K-01 is a 12 bit adc so its 100% well values are Hex16 #FFF0.
Using the bright red photo, it was possible to deduce which wells in the row, column array were red, and following that, which wells were blue and green.
The snapshot of the resulting spreadsheet is below. Note that on the gray card, the red wells are only about half the levels of the green.
Note that the sensor levels for all RGB channnels in both of the well exposed photos are around 8% to 15% of full well capacity.
This is necessary in design so that the camera has 2 or 3 stops (~20dB) of overhead to handle bright spots in the typical images.
(Both of the images I took were about 3.5 stops below saturation, perhaps I could have made slightly higher exposures)
When the camera makes the raster jpg image, it applies the Bayer demosaicing coefficients to the values of the RGB raw well values.
It also applies a gamma correction which boosts the low well values and compresses the high values.
This correction might be 0.45 to comply with ITU standard 709 so the camera jpg image is compatible with various monitors etc.
When the camera displays the histogram, ( I believe) the 3 histograms for the RGB cvhannels are not based on actual sensor well values,
they are based on the resultant pixel levels of the jpg raster image after the coefficients and gamma are applied.
That means, the 3 camera histograms give us an indication of how close to saturation the raster image will be when eventually displayed on the monitor.
The camera histograms do not give a direct indication of how full the sensor wells are.
|
|
Forum: Troubleshooting and Beginner Help
09-17-2015, 10:31 AM
|
| |
Silicon absorption coefficient is lowest for the red filtered wells in the Bayer array, so it follows that red will have the worst signal-to noise ratio and the worst quantization jitter by the adc.
Also the Bayer sensor has a Nyquist on red annd blue that is only half of the green's Nyquist, so aliasing on red and blue will be more prevalent than on green, and sharpness will be worse than green.
When we open a camera raw file in an image processor on our computer , we may or may not be using the demosaicing coefficients provided for that camera by the camera manufacturer.
I read that some of these coefficients in the past have been developed by independent testers who provided the coefficients based on the "best" imagery.
Also when a new camera is released, how do we know what coefficients are being used for it by our older image processing application?
dcraw
https://www.cybercom.net/~dcoffin/dcraw/
In this link is the list showing the main photo image processors that use dcraw along with a list of supported cameras, but you would have to have a version of dcraw after the camera release.
(Note the comments by the author)
Some refs
http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.187.5616&rep=rep1&type=pdf
LinuxPhoto.org
And Holst and Lomheim
|
