When to sharpen?

by Alex Brown 18. November 2012 17:03
Unsharpened original

I have been fretting recently about sharpening, particularly of images being shown on the web, in blogs such as this one. Common wisdom has it that sharpening should be the last thing done to an image, but when images are being hosted by third-party photo services, it can be difficult to control exactly how sharpening is applied.

Flickr’s sharpening

The image at the top of this post is a 640×480 pixel JPEG served directly from Flickr. From an original RAW image, it was cropped and adjusted in Lightroom 4.0, and then uploaded to Flickr at full resolution (6140×4912 pixels) without any sharpening applied. Flickr generated a number of smaller sized images, and automatically applied sharpening to them — what you see at the top of the page is one of these smaller images, exactly as it is when viewed on the front page of my photostream. The effect of the sharpening is quite subtle, but easily visible if you compare it to a 640×512 images made with no sharpening (but just a simple bicubic resize in Photoshop). Mouse over the image below to see the effect of Flickr's sharpening.

Unsharpened resized image; mouse over to compare with Flickr's version

Subtle it may be, but to my eyes it’s not quite right: the urn is fine, and the trees in the distance are fine – but the trees and foliage in the middle distance are oversharpened and rather “etched” and “busy” in appearance, in what is already quite a hight contrast scene with plenty of edges. Some colour richness appears lost.

The effect is more prounced if the original upload to Flickr is sharpened, as then the dreaded sharpening-on-sharpening sin is committed. The image below compares Flickr's image made from the unsharpened full-size original, to one where the original full size image had sharpening applied in Lightroom. Mouse over to see the difference.

If the submission to Flickr is sharpened, Flickr double-sharpens it

Again, the effect is subtle: to me the effect of the double-sharpening is to make the distant trees now seem slightly over-sharpened, and not just those in the middle distance. The sharpening again seems to cause a colour-shift, with even more of a loss of warmth and richness in the receding tree line.

Remedies

And so, the lesson is: if you are concerned about sharpening, be aware that Flickr does it for you. There may be no need to apply sharpening to full-size images uploaded there, if they are intended to be viewed on screen at reduced size (the usual case). The downside of doing this is that users who go on to view the image larger, and especially those who pixel-peep at 100%, will not enjoy an optimally-sharpened image at those sizes.

As an interesting side-note, the fantastic fine-art photography site 1x.com requires its submissions at quite a small size: no more than 950 pixels wide or 850 pixels high. This means it is the photographer’s responsibility to get everything pixel-perfect within those boundaries, safe in the knowledge that no further processing will take place. The downside of course is that viewers with high resolution displays will not enjoy viewing those pictures in their full glory.

For my image, displayed at 640×512 pixels, I prefer to sharpen selectively, after re-sizing. In general I think branches and dense foliage usually need little or no sharpening; and so I apply sharpening mainly to the foreground (the urn, for example). To see the selective sharpening mouse over the image below. Of course there is no way Flickr (or any other service) could have achieved this: the only way to present the image like this is to process it at its final size and then host it as a static resource.

Mouse over to see image with selective sharpening

Far future solutions?

There seems to be a piece of the technology stack missing here, if web images are to be displayed at their best for the increasing variety of screen size and resolution users have. The sharpening required for optimal display on a phone is not the same for as that for optimal display on a large high pixel density screen. Perhaps one solution might be for images to carry (or have associated) metadata hinting at how they should be sharpened at different display sizes, perhaps even conveying information about regions of the image which require different processing …

(Technical note. The image in this post was taken with a Nikon D800. Bicubic resizing was done with Photoshop Elements 8. Everything is in the sRGB colour space.)

Comments are closed

About the author

Alex Brown


Links

Legal

The author's views contained in this weblog are his, and not necessarily of any organisation. Third-party contributions are the responsibility of the contributor.

This weblog’s written content is governed by a Creative Commons Licence.

Creative Commons License     


Bling

Use OpenDNS  

profile for alexbrn at Stack Overflow, Q&A for professional and enthusiast programmers

Quotable

Note that everyone directly involved in the development of ISO standards is a volunteer or funded by outside sponsors. The editors, technical experts, etc., get none of this money. Of course, we must also consider the considerable expense of maintaining offices and executive staff in Geneva. Individual National Bodies are also permitted to sell ISO standards and this money is used to fund their own national standards activities, e.g., pay for offices and executive staff in their capital. But none of this money seems to flow down to the people who makes the standards.

Rob Weir

RecentComments

Comment RSS