Three rolls of Fuji 1600, 7:30 Xtol 1+1. Three rolls Delta 400, 15:30 Xtol 1+1 for ISO 800.
|Standard Resize||Nearest Neighbor||Split|
When photos appear on the web, they both prosper and starve. The range of brightnesses far outstrips anything you could put into a print, rivaling what you can see on a slide. At the same time, the number of perceptable colors can be poor, especially given that the web viewer may have a monitor that's uncalibrated or set up completely differently from the photographer/publisher.
|At a minimum, your monitor should be able to distinguish these individual tones|
Worse, the pictures are small.
One of photography's best qualities is its ability to render lots of small details simultaneously. The web destroys that. A photo may show hundreds of figures, toiling in the mud of a Brazilian gold mine, each laborer with a unique and specific posture, load, and expression. The reduced web thumbnail shows only a dark mass of random arms, five pixels each. Eyelashes, baby peach fuzz, the subtle curl of an eye in either contempt or love... even a full-face portrait can and will suffer from being reduced to a 400-pixel JPEG.
One quality that I like, yet which is almost always eliminated from web photos, is grain. The three photos in the table show some example experiments, trying to hold onto the feeling of a photograph's original grain while still blowing it down to a small, web-happy size.
The original source image was a 35mm negative, Tri-X processed in Rodinal. Lots of clean edges, high contrast, and a fair amount of lovely grain that's abundantly evident in a ~3700×2500, 2820-dpi scan (You can click-through any of the samples to see a 50%-scaled detail).
The first small example image is resized by what I've been using as a "standard" method what I call a "resize/sharpen cascade." Resize first to an even factor of the final size, then resize down in even steps, sharpening at each stage a lot on the large pictures, and progressively less on the smaller versions. The sizes I use most often I've reduced to Photoshop actions, based initially on some mail conversations with Wilfred van der Vegte. The results are sharp, smooth, great for many images but they lose the texture that I love.
The middle image is resized with a dead-simple technique turn off bicubic filtering. Resizing directly to 270×400 with the Photoshop sampling set to "nearest neighbor" preserves the random contrasts and some of the pattern of the original grain, but at the risk of overwhelming the image.
The third example uses a trivial hybrid that's been the most-successful for me so far. I use nearest-neighbor sizing to twice the target size (i.e., 800×540), then size down one more half-scale using the bicubic filter, and then finally a light pass of USM for crispness (around 50%, 1.0 pixels, threshhold zero). While all of the samples lose details that I feel are important to the photo, like the teeth and the faint, tiny hairs on the face, at least this one carries one of the large-print qualities that's important to me but usually unseen in web photos.