1 person found this helpful
Well, as you mentioned, there are R, G and B pixels and there are 2x more greeen pixels than R or B. There are various interpolation methods, but in general, resolution of the photo comes only or at least mostly from the green chanel. So edges, where change in green chanell is small, appear less sharp. Another thing, ACR sharpening is less intensive in dark areas, to avoid noise boosting
About difference between R and B objects, you should check the colors as your camera see them, in the sensor color space (use dcraw for instance), where ratios between the colors are different than in output color space, so if you check your photo for instance, you will see that red letters or background are much more saturated than 'blue'
The worst scenario for bayer sensor is red/blue edge. Several years ago, I made some pictures of red bridge over blue sea, with my 400D (10 Mpix) and older Canon film camera, and elements of the bridge with blue background appear considerably more sharp on the film. But it's onyl extreme case, occuring rarely, except in some white papers for the Foveon sensor
Yes, you've made the point well about the need for the green channel to show fine luminance detail, and to be fair the violet letter on red background in the image above is pretty fuzzy (red and blue data, but little green data).
I've always also suspected, however (and you touched on it with your comment about selective sharpening), is that the red channel may be more noisy and subject to more noise reduction - in Canon cameras at least - because the light levels reaching the red-filtered photosites are somewhat suppressed by the Antialiasing/IR suppression filter (which is blue-green colored), and thus the measurements must already be boosted to compensate.
You are correct
Checked one typical daylight photo (5200K) on 400D, and RGB multipliers, measured by the camera (=WB as shot, as reported by dcraw) are 2.22, 1.00, 1.44.
On my Canon A620 compact, RGB multipliers for the same scene are 1.70, 1.00, 1.60
On the other side, when 400D WB is set to tungsten (3200K), RGB multipliers are 1.50, 1.00, 2.41, so in that case, there is more blue noise
1 person found this helpful
The central problem is the red/blue boundary. As you've already noted above, red/blue boundaries (with weak green) in the native camera coordinates will have lower resolution than other boundaries.
I have not seen any evidence that Canon is performing noise reduction on the red channel any differently than the other two channels.
It is true that the red channel is relatively "dark" compared to the green channel, but this is not the reason for a fuzzier edge.
My last reply to Noel's post was only about higher WB multiplyer for red channel than for green and blue for daylight, which mean, I suppose, that noise in whitebalanced image is higher in red channel. Of course, I don't know the details about noise reduction in Canon (whether on sensor itself or during processing) or in ACR. Only thing I noticed is that ACR sharpening has less effect in the dark tones, like SmartSharpen in PS
However, I also noticed that Clarity tool, which looks to me like USM with larger radius, has also less effect in dark tones, so dark parts of the picture are not 'clarified'. Any reason for this (except because of noise)? I personally preffer ordinary USM with larger radius in PS, instead of Clarity tool for that reason
Hi Vit, yes, you are right that the white-balanced image will generally have more noise in the red & blue channels, for exactly the reason you mention.