What's the best way to resize an image?

August 23, 2005 10:20 AM

What's the absolute best (as in quality) way to resize a digicam image for viewing on the web?

Currently, I have a 5MP Canon point-and-shoot that seems to take altogether decent shots. However, whenever I try reducing them in size for web display—to about, say, 500 pixels in width—I am invariably disappointed in the sharpness of the image.

My current method involves: (1) opening the image in Photoshop, (2) adjusting levels, curves, what have you, (3) reducing the image in size using PS's native function, and (4) applying an Unsharp Mask, usually in the area of 0.4-0.5 pixels, 100%, 0 threshhold.

I have also tried Fred Miranda's Web Presenter Pro, which resizes in increments down to the desired size.

The results are OK, but not on par with what I see elsewhere on the web (cough). I feel like when I first resize the images, they are ridiculously soft. The Unsharp Mask brings it in line *a bit*, but I can never get a nice sharpness: before I can get rid of the softness, I start seeing halos from sharpening too much.

Am I screwing something up? Or just expecting too much from my $400 camera? If the latter, is it a focus problem or a straight up megapixel deficiency?

Here's some sample images I've taken, resized, and sharpened with UM.

posted by deadfather to Computers & Internet (12 answers total)

Well, looking at both sets I'm not sure what you mean. The (cough) sets you linked to don't look sharper than yours. They do, however, have a much shallower depth of field (area that is in focus) and so they have a certain "pop". Shallow DOF is difficult to achieve using a p&s digicam.

So I don't think sharpness is the issue here.
posted by selfnoise at 10:40 AM on August 23, 2005

Best answer:

Many professional users prefer to do Unsharp Masking of color scans on the Lightness channel of LAB mode. Since the Lightness channel has a maximized tonal range and, therefore, presents maximum contrast, the shapening effect, which is based on areas of contrast, is also maximized. Furthermore there is less chance of colored artifacts developing from slight oversharpening while sharpening in the Lightness channel. (Going from RGB to LAB mode and back again does not deteriorate the file at all.) (source)

To be more precise, the venerable Eliot Shepard uses the following steps:

# Select "Lightness" channel, view all channels
# Unsharp mask in Lightness channel (defaults [45%, 0.3, 1], manual adjust)
# Convert to RGB

posted by rafter at 10:54 AM on August 23, 2005

I use the "bicubic sharper" method to downsize images in PS, and I'm happy with it. I've never had a problem with a 500 px image seeming soft; my problem is almost always the opposite (an image seeming sharp at 500 px, but not at full size). But sharpness is not a strictly quantitative thing, so I guess it could be subjective.

I downsize 3000x2000 px images to 400x266 px images with, if any change at all in the sharpness, an apparent increase, using nothing other than Photoshop CS2's "bicubic sharper" option to change the image size.
posted by teece at 11:03 AM on August 23, 2005

Response by poster: I downsize 3000x2000 px images to 400x266 px images with, if any change at all in the sharpness, an apparent increase, using nothing other than Photoshop CS2's "bicubic sharper" option to change the image size.

I'm stuck with v7 for now, which boasts no such feature, unfortunately. (And thanks for the tip, rafter, will definitely try that as well.)
posted by deadfather at 11:08 AM on August 23, 2005

By and large, your daylight samples look pretty sharp to me.

The low-light shot (juniors.jpg) was probably a somewhat long exposure, or shot at a high ISO, which in turn triggered a too-aggressive noise removal algorithm (a common complaint with Canons).

house.jpg looks like bad digital zoom (turn it off in settings and live with what the camera can do optically).

In Photoshop Image >> Image Size..., make sure you're using the "bicubic" resample option.

A more fiddly tip would be to resize only to power-of-two fractions (1/2, 1/4, 1/8, etc.) of the raw size. There might be marginally less rounding error/aliasing that way.

Remember that the clarity and dot pitch of your monitor also influence apparent sharpness.

And last, yes, a nice prime (non-zoom) lens on a 35mm camera is always going to look a little better than the fixed zoom on a consumer digicam.
posted by eshepard at 11:24 AM on August 23, 2005

Something I've found that helps when resizing large images is to do it in multiple resizes. Going from 3000px to 500px can cause some jagged edges, something I noticed in your arch photo. But resizing in stages (from 3000 to 2000, to 1000, to 500 for example) lessens that effect, after which you can sharpen.
posted by rhapsodie at 11:48 AM on August 23, 2005

When resizing an image in Photoshop I use Canvas Size to keep the pixel dimensions even on both sides. Adding or subtracting a pixel this way avoids the rounding error/aliasing problems eshepard mentioned.
posted by tut21 at 12:30 PM on August 23, 2005

Best answer: I sharpen twice. Once before I resize, then again after the resizing is complete. So for a digital painting that's about 5000 pixels on one side, I'll unsharp mask at .6 to 1, resize to web size (800 or 900 pixels on its' longest side), then unsharp mask at .4 or so.

Of course, most digi cams that I've used leave a huge amount of deeply ugly noise in the blue channel, which wreaks havoc with my digital painting sharpening method. So before I sharpen a digital photo, I switch to LAB mode and fix the noise issue there by blurring the A and B channels a bit.
posted by xyzzy at 12:40 PM on August 23, 2005

What everyone said.

Another thought: I think there's a human factors aspect to it, as well. How are you assesing sharpness? It's something we can all get over stressed about. I've found myself peering at pictures up close (too close) to the screen while fussing with USM controls, trying to squeeze that very last possible iota of sharpness from an image for web presentation. Then realizing that at such a close viewing distance from a CRT no photo is going to appear perfectly sharp. Leaning back in my chair to a sensible viewing distance... "Hey, actually, that doesn't look half bad!"
posted by normy at 1:32 PM on August 23, 2005

I've never noticed anything but an increase of sharpness when sizing down. Are the full size images sharp? That makes the biggest difference.
posted by Nothing at 3:04 PM on August 23, 2005

Response by poster: I recommend the LAB color trick to everyone. It really helps out. Thanks all.

I'm still agnostic on resizing in steps (Fred Miranda method). I can't tell a difference, except occasionally I lose some detail.
posted by deadfather at 9:10 PM on August 23, 2005

Try sharpening before reducing the size of the image. When reducing it shouldn't even make too much of a difference to sharpen at all. Probably the best thing to do to spice up the image is to improve the contrast. I find that makes them pop more.

TTTT, your images look fine.
posted by JJ86 at 4:00 AM on August 24, 2005