Untitled Document

It seems to be a commonplace that one should resize and then finally sharpen. I don't think this is quite right, especially if the resizing is going to make the image much smaller (say for display in a browser.) The problem is that USM gives a lot less fine control over radius after such a resize. You get nothing at .1 and too much at .2 (say). It's even worse for HIRALOAM.

So I propose that the actual right thing is to try to view the iamge at some reasonable approximation of it's intended viewing size (and some even multiple of it's current size) while sharpening.

Does this sound right?

--John Ruttenberg

___________________________________________________________________________

John,

Bruce Fraser has written a whole book on this subject called "Real World Image Sharpening with Adobe Photoshop CS2". Highly recommended. He describes in detail the merits of a multi-stage sharpening workflow in which the relationship between resolution and sharpening is also dealt-with. In that approach, Stage 3 sharpening occurs after the image resolution is finally decided. Furthermore, with sharpening, what you see on your monitor is not necessarily a reliable guide to how the sharpening will appear in print or on someone else's monitor. You also need to remember that at magnifications other than 25%, 50%, 75% and 100% in Photoshop the image is aliased, which further detracts from reliability of wysiwyg.

Mark Segal

___________________________________________________________________________

It has been my experience that USM does not work well when trying to sharpen image for internet viewing, i.e. 72 dpi.

When formatting a file for web I do the following:

1. Resize the image using >mage>Image size, check resampling, and Bicubic Sharper

2. Define the size you want the image to appear as and then set the resolution to 72 dpi

3. Once the image is at the designated size at 72 dpi, view it at 100%

4. Filter>Sharpen>Sharpen

This brings the image - even thumbnails alive and viewable in the browser. Most viewers are not "color geeks", and few know what a color corrected monitor is, so seeing an image where they can discern what it is - is more important than color quality.

Fine Photography by

Michael Demyan

www.mikedemyan.com

610-758-9769

___________________________________________________________________________

If we're talking sharpening for the web, I find this difficult to pull-off successfully, as I'm very sensitive to over-sharpening images. I find that if I only do "Capture Sharpening" (using PK Sharpener), then prep the image for the web, not only that it doesn't need any more sharpening, but also looks more natural. PK Sharpener also has web output sharpeners, by I find these also can be aggressive. Monitors have a way of making even slightly over-sharpened images look crinkly, so it best to go very easy.

Mark Segal

___________________________________________________________________________

John Ruttenberg wrote:

Hi John, I agree that this is the common legacy workflow - this used to be the case, but with Bicubic Sharper some choose to simplify their workflow and do it all in one step.

I presume you mean sharpening to restore the appearance of sharpness in the resampling down process, rather than for output.

For many images/content and resampling sizes, it may not matter too much - but then there are other more problematic images with repeating patterns or fine detail that may have to be handled in a more creative way. This may include sharpening and then reducing using multiple steps (sharpening before each res down), or combinations of resampling methods merged, or overly emphasising minor detail before the resolution drop etc. Print has more flexibility for fine repeating patterns and one can try more tricks as it is not at 1:1 like monitor display is usually viewed.

This is an important subject John and I think most just use their favourite one size fits all method and only look further when a particular image has negative issues - which are usually dependent on the image, method and resolutions.

And *then* resample down?

Perhaps comparing Bicubic softer to Bicubic sharper and or other methods.

And then decide if some further sharpening is required?

Even for a monitor, what looks soft on a CRT at 1:1 may look too sharp on an LCD, so one would have to decide on the major audience etc.

So many variables!

Regards,

Stephen Marsh.

___________________________________________________________________________

Back in the day when people used modems to surf the internet there was a lot of thought given to reducing page size. There was a page that advocated this for thumbnails.

First, blur your image. Intensify the colors. Reduce to thumbnail size. Sharpen.

This gave a remarkable thumbnail. I know because people remarked! I never understood what was going on but your post reminded me of it.

___________________________________________________________________________

Gene Palmiter wrote:

How times change!

Sharpening adds to JPEG file size, but may be a necessary evil. One can slightly reduce file size by blurring the colour component and not the luminance, but luminance blurring provides the biggest saving for JPEGs.

Perhaps what was going on is what is mentioned here:

http: //www.cambridgeincolour.com/tutorials/image-resize-for-web.htm

Some images or content may benefit from a blur before resampling down.

Some images (often logos or graphics or CG) may benefit from sharpening before sizing down.

Some may need combinations of both.

Despite upsizing getting more press, I have always thought that downsizing files should also gain wide testing and commentary.

Regards,

Stephen Marsh.

___________________________________________________________________________

on 3/4/07 9:44 PM, Michael Demyanwrote:

A site that sells fashion items would find this approach costly (returns). In this case without "color quality" no one would no what they're looking at either.

Lee Clawson

2/\V/\7 Studio

___________________________________________________________________________

A fashion site that does not place a disclaimer regarding displayed color vs. actual color will have costly returns.

The original fabric/item is photographed - that image is not always the exact color of the item (we try to get it as close as possible) it is then converted to sRGB and placed on the web to be viewed by the public - 99% of whom do not have color calibrated monitors.

Printed catalogs also have the same disclaimer - since photographed items - converted to CMYK color space and printed on a variety of paper stock - can not in many cased accurately display the true color of that item.

An image placed on the web should have colors "as close as possible" to the original but should be sharp enough to be able to discern the texture/detail of the item.

This site is devoted to color. We all know it is subjective and we try to make it as close as we can - to how WE see it.

Fabric/Wallpaper/Rug manufactures will not guarantee color matching from one run to another. How close do we as image makers have to be?

Mike

Fine Photography by

Michael Demyan

www.mikedemyan.com

610-758-9769

___________________________________________________________________________

Mike,

My point was/is that unlike your example I have many end uses where we go with color above all other concerns. I was trying to show that other industries have the same concerns.

Lee Clawson

2/\V/\7 Studio

___________________________________________________________________________

Hi Lee:

I can agree with your reasoning - as long as all of your/their end users have color calibrated monitors as yours/theirs, and understand the color gamut shifts between color spaces. I sympathize with the other industries.

Color is subjective. Unless it is measured using calibrated instruments under controlled lighting, it is a crapshoot. Industries recognizing this usually provide special lighting booths - as in the retail sector, where the client can view the product samples under controlled lighting. Some vendors will suggest bringing a sample to the clients environment and displaying it there, under the clients lighting conditions.

The end result - the client must be happy when viewing the product under their lighting conditions.

Mike

Fine Photography by

Michael Demyan

www.mikedemyan.com

610-758-9769

___________________________________________________________________________

on 3/6/07 3:05 PM, Michael Demyan wrote:

Mike,

I'm not sure I agree completely with you but this (above) statement I whole heartily agree with. We still have an in-store lighting fixture for a client who demands their screen-printed signage must match the spec'd color standards.

Lee Clawson

2/\V/\7 Studio

___________________________________________________________________________

Matching logo or ID colors is always a priority. THIS IS the identity of the client. I will agree with you on this point. You can't make a Ritz cracker box look orange and sky blue.

What I was referring to was the colors of diverse products as placed on the web or in a catalog. We try to get them as close as the technology will allow.

Fine Photography by

Michael Demyan

www.mikedemyan.com

610-758-9769

___________________________________________________________________________

Okay Stephen,

Let's talk about downsizing and USM.

With the invention of the R2400 K3 printer I decided it was time to wade through my collection of b&w negatives that I shot some 20 years ago and print/reprint them with this new technology. With the final purpose being a complete and consistant portfolio of my early work.

WOW... but I won't go there because I could spend an hour talking about how wonderful editing and printing b&w photos this new way is compared to the darkroom.

But basically this is what I did. I scanned all my negatives at full resolution (4000ppi) on a Coolscan 9000 with the knowledge that I would downsample to make the final print at whatever size I decided in the future. Some quick tests gave me confidence that, compared to scanning for each intended output size, I wasn't doing anything that I would regret.

I made the first prints at 11X17 @300ppi inches which is actually pretty close to the maximum resolution of my scanner so not much downsampling was required.

The second prints were sized a 6.5X10 inches @300ppi. The big decision I faced was whether to take the final 11X17" photo, with all its corrections and USMing and just downsample it (using bicubic sharper) to the smaller size or take the unsharpened version of that photo, downsize it and then resharpen with USM at the smaller size.

I tried both methods and really couldn't see an advantage to resharpening the smaller print. But I'll admit that comparing the two methods was very difficult.

First, I learned that resharpening the smaller print with the same USM settings as the larger print isn't a good idea. So how can you be sure that the new USM settings for the smaller print are doing exactly the same thing as with the larger print at it's settings?

Second, how do you take a sharpened photo, downsize it to 50% its original size and compare them on a computer monitor? If I tried to enlarge the smaller image I ran into pixelation problems. And if I tried to view both at 100% the smaller image was too small to compare it to the larger photo. What actually seemed to work best was to size the images so they were a comparible size on screen and then back away from the monitor until I could tell if there were any strange tonal changes or artifacts in the downsampled image.

So finally, as is the philosophy of this group I decided to use the final prints as the judge. And, once again, I couldn't tell them apart... I think. Again, the smaller print naturally appears sharper anyway with less grain irregardless of which method I used to downsize it.

So Stephen, I've been intrigued in the past about you wanting to talk about this subject but I wonder if there really is an issue to worry about when it come to downsampling.

I will add that my early work in b&w does not meet the 'f64' group's criterion for highly detailed studies of intricate subjects. And in fact are very dark and moody in nature so they may not have been the best negatives to use for such a test. But they would prove that, at least for me, downsampling is not an issue.

And one final note, if I take the final sharpened photo (in either size) and downsample it to a size suitable for viewing on the web I find no reason to sharpen or resharpen the photo either. In fact, with any of my photos, new or old, I have never even given special sharpening for web use a thought... I never do it.

Murray DeJager

___________________________________________________________________________

Murray,

Could you tell us what, if anything, happens with very wide density range color images.

Lee Clawson

2/\V/\7 Studio

___________________________________________________________________________

Hi Lee,

The only caveat on my present tests are that, so far, I've only been printing B&W at these sizes and with this method of downsizing.

I've returned to photography after a long absence so my experiences are all fairly new. But after studying Dan's books and methods I recently decided I was ready to start making 'final' prints of some of my early work.

Starting with the B&W photos seemed to be the best plan. And now I definitely think it was the right plan. Applying curves, noise reduction and sharpening on only one black channel has been much more relaxing.

I have yet, however, gotten to my color photographs from the past or my present color photographs and corrected them with a 'final, perfect' print in mind. I have done a lot playing though. I don't know yet if what I've learned with these b&w photos, especially with USM, will transfer directly to color work with multiple channels.

Not all my b&w photos that I've printed to date are lacking full tonality though. But even those with a complete range of tones have not shown any negative effects from downsampling from the larger 11X17" size to the smaller 6.5X10" size.

In Dan's latest book, I believe, he suggests that sharpening isn't a major issue with smaller prints. I'm not sure what Dan considers small but I'm guessing that he's referring to prints in the 4X6 - 16X20" range. Where sharpening becomes very important is in the 24X30" range and up. Forgive me Dan if I'm wrong. But anyway, from my limited experiece so far I would say I agree with him.

My past photos are turning out far better today than when I worked on them in the past so I'm very happy with the results. Once they're done I will sit back down and study this subject much more. Rereading Bruce Fraser's book on sharpening and finishing Dan's book will be a priority before working on my present color photos, at least with the goal of perfection in mind.

But basically any concerns about downsampling just have not surfaced for me. I do have a feeling, however, that like the conversations on uprezing a lot may depend on the subject matter in the photograph.

Murray DeJager

___________________________________________________________________________

John Ruttenberg writes,

I have not heard this as a commonplace--quite the opposite. Downsizing the file may well degrade the edges that USM needs to find. So, in some images at least, conventional USM--not hiraloam--may need to be done before downsizing.

Even assuming that a preliminary conventional sharpen appears to be the best technical approach, the subsequent possibilities are almost unlimited. Do we try to sharpen hiraloam also before downsizing? I can see theoretical possibilities for it being good--or bad. After downsizing, do we sharpen conventionally again, or do we try to avoid it? Does the general rule that darkening halos are more tolerable than lightening ones really still apply if we downsize? Is there any case for downsizing halfway, sharpening again, and then choosing the final file size?

We could probably find out some tentative answers if somebody were willing to spend about two weeks doing nothing else but testing images. Evaluating sharpening methods is trickier than almost any other area of image processing because so many different factors are in play in specific images.

I certainly spent a full two weeks trying different methods for PP5E, using a suite of around 100 images--and still I got some points wrong. And that testing didn't contemplate downsizing.

Same way with the question of increasing resolution. There are a good dozen ways of doing it. Sometimes you get better results with one, sometimes another. The only way to get a definitive answer is for somebody to spend more than a week studying it.

Last time I did this was several years ago, testing two third-party programs against simple Bicubic in Photoshop. I concluded that there was almost no difference at less than doubling the resolution, and thereafter the programs did better if there were pronounced edges in the original and Photoshop better when there were large smooth areas such as faces.

Today, we have lots more options--two new algorithms in Photoshop, plus various schemes for partial upsizing with or without sharpening along the way. Sometimes one method gets a noticeably better result than another, although we rarely see really dramatic differences. So, some people are more comfortable with one way than another, but nobody really has definitive answers.

I don't think we will *ever* know, either, because the gain of knowing this information doesn't compensate for the pain of uncovering it, particularly when it's unlikely to provoke anything more than minor improvements. Plus, the differences only show up when the up- or down-sizing is relatively large. Anybody who has to do this frequently probably has many more workflow issues that could be more profitably addressed, and so is better off winging it based on his general knowledge of imaging.

No. Unless you are specifically looking at a 50% or 25% reduction, Photoshop's preview is not going to be accurate enough for serious evaluation--the way it displays a large file at low magnification is not going to be the same as displaying the file at the same screen size after it has literally been downsized.

My approach: as a rule, it's a bad idea to sharpen twice at similar Radius settings, so I dislike two conventional sharpens to the same file. Also, I don't know enough to say whether hiraloam USM works better at higher or lower resolution; I assume it's safer to do it after downsizing.

Therefore, I do the conventional USM before downsizing, on a separate layer. I always use 500% Amount (Radius and Threshold may vary). The idea is to have something too sharp rather than not sharp enough after the downsizing.

If the image is not particularly important, I then blend the lower (unsharpened) into the top at 50% opacity, Darken mode, to cut the light haloing. If the image *is* important, instead of the above, a copy of the bottom layer goes on top, sandwiching the sharpened copy in between. This top layer is set to 50% Darken, but I reserve the right to change that after downsizing.

Dan Margulis

___________________________________________________________________________

Hi Dan,

Do you ever do anything that resembles a "multi-pass" sharpening of images, taking into account input devices and output devices?

Am I correct in assuming that you always use 500% Amount because this setting only controls the degree of whiteness and blackness of the sharpening halos. And adjusting the opacity of the layer on which you do your sharpening is basically the same thing? And one could get exactly the same results by using the Amount slider instead?

Like the 500% Amount/Opacity setting the above just allows you some latter control of the sharpening halos, correct?

Assuming my above questions are correct, then with CS3 I've read that by using Smart Filters you'll be able to go back and reset USM settings at anytime after the original settings have been made with ease. Then you would have control of all the USM setting and wouldn't have to use separate sharpening layers with an Amount setting or 500% and a lower opacity setting to adjust the brightness of the halos, correct?

Of course to adjust the white and black halos separately one still needs a work-around.

Murray DeJager

___________________________________________________________________________

Murray writes,

Double sharpening is only effective when the Radius settings are significantly different. Otherwise, the artifacts of the first sharpen are emphasized disagreeably. It would always be better to do one larger conventional sharpen than two smaller ones.

The impact of output conditions is minuscule in comparison to getting the correct settings for the character of the image.

Yes, this is simply a matter of workflow choice. I used to do it with the Amount slider, now I find it it is more comfortable in this (and several other correction contexts, such as LAB curves) to overdo the desired correction at first and then adjust it downward with layer opacity and/or Blend If options.

Obviously the 500% Amount is only for conventional sharpening. When I do hiraloam I *start* at 500% to evaluate Radius, but then reduce it sharply, usually to around 50-75%, before blending to reduce lightening halos, and then a further reduction of layer opacity. As long as the top layer is sharper than you could possibly want (and 75% with a very wide Radius qualifies) it's counterproductive to go much higher.

Yes.

I haven't worked with the CS3 beta and so don't know whether there is a workaround, but I suspect that it doesn't help for sharpening, because as I understand it there would be no convenient way to control lightening and darkening independently with the current Smart Filters implementation. We'll see when the product is released.

Dan Margulis