I've been processing a few photos of birds taken at sunset recently. Just wanted to make sure what I do is in accordance with best practice.
I act on the principle that the value "255" should not appear in the jpeg file on the subject in any of the three channels, except in specular highlights (e.g., droplets of water on a duck). Thus, if I see any pixel on the subject at 255 in any of the channels, I use the top slider in the raw side of DPP to darken the photo until the 255 value doesn't appear on the subject. (In practice, red is almost always the channel that blows.) I then go to the RGB side of DPP and apply a suitable brightening curve "to taste" to make the subject not appear overly dark. The curve always passes at a reasonable angle through the top right corner of the histogram box in order not to reblow any pixels. (Any other implementation of curves would also work in this process, I would imagine.)
I've been getting photos of birds at sunset that needed 1 1/2 stops of darkening to get rid of the 255's in the red channel. This happens even though the photo as a whole looks reasonably exposed (or even underexposed). I assume this is because the incident light is reddish. The resulting histograms after darkening have relatively little content in the top stop or two, no surprise. I have started to tolerate a few dozen R=255 pixels to avoid further darkening.
Similar effects have occurred in ordinary sunlight on the epaulettes of male red-winged blackbirds, by the way.
Hi Flavio. Many times the issue is not overexposure, but over-saturation of the red channel. As I understand it to prevent the reds from clipping you are adjusting the exposure of the complete photo...hence resulting in way underexposed results. Try this: once you've got proper exposure selectively desaturate the red channel until it is no longer clipped. Here are a couple of must reads when it comes to the specific problem with reds and many digital cameras:
Try lowering the CONSTANT in the Channel Mixer Adjustment layer with red channel selected. Many times this will also result in less banding issues than other methods.