You can use statistical blending to render high dynamic range images, to reduce noise, or to create multiple exposure effects in Photoshop (CS4 extended edition). All these techniques require that you have multiple images to start with. For HDR images, you need different exposures of the same subject from a stationary viewpoint. The same is true for noise reduction, only the exposures should be identical. Multiple exposure effects can use any number of different images, all of the same dimensions. In each case, you start by stacking all the images into layers, selecting all layers, and converting them into a smart object. Then you use the Layer > Smart Objects > Stack Mode functions Mean or Median to create a statistical combination of all the images, which you can rasterize. Details after the break.
First you’ll need to load some files into a stack. You do this from Bridge by selecting the files and then choosing Tools > Photoshop > Load Files into Photoshop Layers… from the menu. If you do this with DNG files, they will be rendered at 16 bits in the stack (if you have Camera Raw configured to deliver 16-bit files). In Photoshop, use the File > Scripts > Load Files into Stack… command. The Tools > Photoshop > Photomerge.. command in Bridge or the File > Automate > Photomerge… command in Photoshop are useful alternatives for stacking images. Use the Auto setting and be sure to uncheck Blend Images Together. Auto only works if the images are all basically the same, as they would be if noise removal is your goal.
In Photoshop now, if you haven’t aligned the images, you should do so. Then choose Select > All Layers to select all the layers, and turn them into a smart object with the Layer > Smart Objects > Convert to Smart Object command. After that, try out the Layer > Smart Objects > Stack Mode > Mean and Layer > Smart Objects > Stack Mode > Median commands. The former provides the average of all the stacked pixels, the latter the middle value within the range of pixel values at each pixel. Both are effective at diminishing the effects of noise: assuming noise varies randomly around a central value, taking several images and averaging them will reduce noise. You can use this strategy for photographing low light situations, for example. Some cameras have settings to accomplish this, too.
Finally, you’ll probably want to rasterize the image. It may be worthwhile to rasterize to higher bit depth first, and then downsample–at least in theory, if you convert the individual images before rasterizing, they should produce more “fractional” values within the range, even if the range itself doesn’t change (entropy works here, if not on CSI). This provides more headroom for histogram stretching. OTOH, most of the value in the technique comes from the statistical operation itself, rather than from post-processing as an HDR, for example. At the very least, it’s an interesting alternative to Photoshop’s HDR tools, very useful for noise reduction, and a good source of multiple exposure effects. Indeed, these effects, in the hands of an artist who gives them his full attention, can be significant, as our colleague Jason Salavon has shown. Jason rolls his own code, of course, so don’t expect Photoshop to do what Jason does. Still, with a little plundering of the Yorck Project’s holding of Frans Hals you might try to imitate (sincerest flattery) Jason’s take on Northern portraiture.
More information: See Adobe’s Help for image stacks.