Down-sampling examples.


On this page, you can find examples of a large original image that was down-sampled for monitor display, using different programs/methods. The purpose of the exercise is to compare the degree of effectiveness of the different anti-aliasing fitering methods as used in several popular programs.

Less effective low-pass prefiltering methods can cause artifacts in down-sampled images because detail that is resized to smaller than one pixel will be misrepresented by larger aliases. Especially regular pattern detail will exhibit visible low spatial frequency patterns, unless the image is properly pre-filtered. These patterns can be very distracting.

Important to realize is that also non-regular patterns will aliase. That can result in unnatural grittiness or semi-structure which can change the appearance of texture, unless proper precautions are taken. It can e.g. change the appearance of leather into cardboard, which is not good if you are making an illustration for a fashion or furniture brochure/web-site.

The original image used for this page was a scanned 35mm color negative film frame that I shot as a practice target for perspective control. It was a handheld shot with an average quality 24mm wide angle lens, and the 5400ppi scan was full frame, including a bit of the filmholder edge. This resulted in an original image size of 7696x5096 pixels (39.2 MP), which is obviously too large in size to make available in a web-link. When printed at 300 ppi the result would have measured 25.7x17.0 inch or 65.2x43.1 cm, and the on-screen size is 3-4x larger, so expect to see graininess in the original size crops.

Below are the 7696/600 (= factor 12.8x) down-sampled images (the JPEG conversions of the resized original were made with Photoshop CS in highest quality), but first I'd like to draw attention to a few interesting image areas, by using several crops from the original full size image.

SunScreen (79K)
Pay special attention, in all downsampled images, to the vertically red-white striped sun-screens of the white building at the lower left of the full image. There are 11 narrow vertical red stripes in the righthand screen facing the camera. In the original image above they are well defined, but in many of the down-sampled images they are represented as a few fat stripes that bend in the wrong direction, thus misrepresenting the shape and structure.

Balcony (110K)
On the main building look for the 1 pixel separation of the short balcony pillars, at the left above the main entrance.

Pole (33K) RoofTiles (174K)

Also notice the, in some images, seemingly stair-stepped metal pole on the highest tower and on the traffic sign in the left foreground. In some images, the wind vane on top of the entrance section seems to hover in mid air.
Wall-Vent (49K) BrickWall (90K)

Check for non-existing arching patterns in the roof tiles, and in the brick walls, especially visible on the lower right-side wall of the building and look at the direction of the 5 slits in the air-vent in that wall. Also the street bricks on the right side can acquire strange coarse patterns.

Now that you know what to look for, critically inspect the down-sampled results below. You may want to download the images you prefer, and inspect them by zooming in. Also check those zooms for colored edge artifacts, due to differences in chrominance sharpening.
TownHall-IrfanView-Sharpen (356K)
The image above was downsampled with IrfanView 3.91. The result has some visible aliasing (e.g. the red-white sun-screens and the right brick wall). The image initially also looked a bit soft, so the version shown here had also some IrfanView sharpening effect (set to 20) applied. The image features have color artifacts (e.g. the pole on the peaked roof), but the result doesn't look too bad for a free program.

TownHall{Q307} (333K)
The image above was downsampled with Qimage Version 2004.307. The result shows significant amounts of aliasing artifacts (esp. walls/roof/street), and the image could use some additional sharpening although that would also sharpen some of the artifacts. If you zoom in on the image, there are also color artifacts visible along high contrast edges.

TownHall{Q310low} (338K)
Starting with Qimage Version 309, and shortly after it was further improved by Version 310, its current anti-aliasing behavior is much better, and user adjustable. The Image above was down-sampled with Qimage Version 2004.310, with a "Low" anti-aliasing setting. Although better in default sharpnening (too much for my taste), there are also still a lot of aliasing artifacts visible. These are especially distracting on the brick wall and the street on the right side and they are made more prominent by the sharpening.

TownHall{Q310med} (284K)
The image above was down-sampled with the improved Qimage Version 2004.310, with "Medium" (default) anti-aliasing selected for down-sampling. Now the aliasing artifacts are further reduced, and practically invisible. The overall result is much better than previous versions could produce, and the automatic sharpening speeds up the workflow by eliminating an additional step. The result is acceptable in terms of default sharpening, although on the soft side and with visible halo. As it is implemented, it is the program's best general choice for subjects where aliasing may rear its ugly head, although the default sharpening could be improved.

TownHall{Q310high} (243K)
The image above was down-sampled with Qimage Version 2004.310, with "High" anti-aliasing selected for down-sampling. Now all the aliasing artifacts are prevented, but at the cost of softening the image too much.

TownHall-BCsharper (90K)
The image above was downsampled with Photoshop CS, using the Adobe recommended "Bicubic sharper" resampling option. The result has a little visible aliasing, but also looks a bit soft. The image features have some sharpening halo/ringing (e.g. roof/sky edge) which makes it difficult to re-sharpen without making the halo more visible.

TownHall-BC-USM (293K)
The image above was downsampled with Photoshop CS, using the standard "Bicubic" resampling option. The result has no obvious aliasing artifacts, but it did initially look a bit soft, so the version shown here had some additional small radius luminosity USM applied.

TownHall-IMlanczos (283K)
The image above was downsampled with ImageMagick Version 6.0.1, with only Lanczos filtering and anti-aliasing selected. There are no visible traces of aliasing, but the image could use some additional sharpening.

TownHall-IMlanczos-USM (306K)
The image above was downsampled with ImageMagick Version 6.0.1, with Lanczos filtering and anti-aliasing selected. In an attempt to approximately match the amount of sharpening done in Photoshop, some PS USM sharpening was applied. Although the amount of sharpening is a matter of taste, the lack of aliasing artifacts produces the cleanest, most natural looking image of them all.

If you have left Javascript enabled in your browser, then you can switch here between the same images, superimposed onto each other for a more accurate comparison.

townhall
Select image: IrfanView 3.91 IrfanView 3.91 + sharpen (20)
Qimage Pre 309 QI 310 Low-AA QI 310 Med-AA
QI 310 High-AA Photoshop "sharper"
Photoshop "bicubic" Photoshop "bicubic"+USM
ImageMagick Lanczos ImageMagick Lanczos+USM


Top of page


You can reach me at this e-mail address with questions, remarks, and suggestions about this page.


Designed by Bart van der Wolf Latest revision: 31-January-2017
Copyright © 2004,2017, Bart van der Wolf