Is it possible to sharpen the highlights and blur the shadows?

Going through some of my latest high rez B/W scans from Delta and Tri-

X got me thinking along a track to help smooth out the grain closer

to the way color films scan.

 

With color film it's already second nature for me to go into the blue

channel and apply some gaussian blur, or to only apply USM to the red

and green channels. That's an old one.

 

For conventional B/W films though it would seem than if you applied

USM or gaussian blur across a decreasing or increasing density range

you could better smooth out the granularity that occurs with

conventional B/W films. It's no big deal in a smaller prints, but in

larger digital prints I find it distracting. The irony being the

better the scan the more the grain.

 

For instance, if you only wanted to apply USM to lighter regions of

an image or gaussian blur to the denser ones what is the easiest way

to accomplish this? Or, is this an already existing tool I'm just

missing, or, has somebody already tried this and found it a waste of

time???

I might be oversimplifying your question here Scott, but if I am not, couldn't you just use the Color Range tool (under Select menu). In B&W it works to select regions of similar denseness. I would think that would work quite well, as would the magic wand tool with continuity off and an appropriate tolerance.

 

unfortunately, it would be incredibly difficult to keep the regions separate. It would be easy to create two selections with one containing lighter and another containing darker portions, encompassing the entire image. However this would combing together regions that probably don't belong together.

It's quite possible (I just tried it before typing this up). Bruce Fraser has an <a href="http://www.creativepro.com/story/feature/12189-1.html">article</a> on using layer masks for sharpening. Now, if you create another layer, invert the layer mask, and substitute "blur" for "sharpen" at the final step, you can have sharp highlights and blurred shadows. If anyone is confused by the article I can write a tutorial.

<p>

Sometimes you can have your cake and eat it too.

I knew layers would come in to this somewhere.

That Fraser method seems a lot of kerfuffle for not-a-lot of result, to me. I accuse him of 'cheating' on his admittedly bad comparison example, by allowing what appear to be JPEG artefacts to be sharpened, as well as simply overdoing the sharpening altogether. Anyone can produce a badly oversharpened example of what NOT to do.<p>Anyhow, back to the thing in hand: I think that's a really good idea you've had there Scott!<br>Creating a dupe of the original, and then boosting the contrast and varying the brightness, would create a basic mask for either highlight or shadow selective blurring. It should also be possible to use the 'select> colour range' tool to create masks for other specific areas as well.<p>I've simply applied blurring to selected areas like backgrounds or skies, in the past, but the edges are fiddly, and never really satisfactory. An unsharp masking layer would definitely be a better solution. Thanks Scott.

What I'm thinking about here in more specifics is applying standard USM or a Gaussian blur, but the threshold or strength of the filter is directly proportional to the simple combined RGB value. Darker vs lighter areas would recieve different degrees of effect depending on the strength of preference. I suppose this would work in each color channel as well, but for now I'm thinking about simple grey scale mode.

 

Yeah, a contrast/layer mask would likely accomplish a similiar result, but this seems conceptually simple enough that there might be a more direct way to apply it along a progressive or linear density curve.

<i>What I'm thinking about here in more specifics is applying standard USM or a Gaussian blur, but the threshold or strength of the filter is directly proportional to the simple combined RGB value.</i>

<p>This is exactly the effect of the layer mask, except it is even more flexible. The article goes into great depth (see <a href="http://www.creativepro.com/story/feature/12333.html">summary</a>), but it really boils down to:

<ol>

<li>Create a channel of the desaturated image using Image>Calculations (Result=New Channel, hereafter "Alpha 1")

<li>Duplicate the background twice; one for the highlights, one for the shadows. Add a layer mask to each one.

<li>In the Highlights layer mask, run Image>Apply Image and select Invert to copy "Alpha 1" to the layer mask. (Engineers: You have created a highpass filter).

<li>In the Shadows layer mask, run Image>Apply Image to copy "Alpha 1" to the layer mask. (Engineers: You have created a lowpass filter).

<li>Using the two new layers (not on the masks), blur the shadows and sharpen the highlights.

<li>Optional: Manipulate the intensity of the filters by changing the layer opacity, or using levels or curves on the layer mask.

</ol>

I've done this on color images. I duplicated the background layer and added a mask which was the desaturated version of the image. Blur the mask because you don't want hard transitions where there is actual high contrast fine detail (rather than lower contrast noise). Unsharp mask the top layer (the one with the mask). Use curves to tune the mask. Works great for scans with lots of noise in the shadows.

Emre: Those highlight and shadow masks that you describe are not the same thing as high and low-pass filters.<p>High and low-pass filtering, in terms of image data, refer to the spatial frequencies contained within the image. These spatial frequencies are the variations of brightness with distance throughout the image. They can be thought of as the line pairs per millimetre of each part of the image data, if you like.<p>Ideally, we would like to apply some spatial filtering in order to attenuate only the grain frequencies of the image. But we'd need to make a Fourier transform of the image data first.<br>Then we could filter out just those frequencies associated with the film grain, and finally re-transform our Fourier series back into a recognisable image again.<br>That would be ideal, but unfortunately we're stuck with doing some fairly crude pixel-by-pixel maths on the image.

I must concede, you are right. However, if you apply "Find Edges", it does indeed become a filter (Fraser mentions this too).

In fact, if you are feeling somewhat bored, you could independently apply filters to the shadows, highlights, low frequencies, and high frequencies. Talk about flexibility!

Scott,<BR>

<BR>

This should be an easy process using the "Picture Window" software from <I>Digital Light & Color</I>. One of the optional masking techniques in PW is to mask based upon brightness range. This makes it a snap to mask an area such as the sky, which is probably the most common area that you would wish to separate out from the rest of the picture. Occasionally, this brightness mask will pick up a few areas outside of what you desire, but this can easily be fixed by applying a simple subtractive brush mask to those areas and producing a combined mask with NOT logic. Once you have the correct area masked you can then perform USM, or Gaussian Blur, or any other desired operation, with the confidence that it will be limited to the portions of the image that you desire.<BR>

<BR>

With best wishes,<BR>

- Tom -

Scott-

 

This seems like an easy question, so I may be missing your point. Anyway, you�ve answered so many 'easy' questions for everyone else that I�ll try...

 

One way would be to go into the channels palette and hit the �Load Channel as Selection� button with either the Gray or RGB channel selected (depending on whether the image is in Grayscale or RGB mode). Then, invert (and probably feather) the selection and run the Gaussian Blur filter. This applies the blur in increasing strength to the increasingly darker areas of the image.

 

To do the USM step, load the channel as a selection again (and, again, probably feather it) and run the USM. This will apply the USM in increasing strength to the increasingly lighter areas of the image.

 

As others have suggested, there is probably a good way to do this with layers. However, with high res scans destined for large prints, it probably isn't worth tripling your file size for marginal (if any) gains. (I assume you've got the original scan saved/archived elsewhere).

 

Hope this helps.

 

-Bill

I did something like that with my EdgeEnhancer plug-in... When a sharpening convolution kernel is applied to a neighborhood and added back to the original image, you are doing one of the following:

 

1) making it darker (especially if it a dark edge)

2) making it lighter (light edge)

3) no change (uniform area)

 

If we decide to ONLY add the dark enhancements or the light enhancements, you can do half of what he wanted to do .. sharpen the highlights.

 

(warning - marketing speak follows)

 

Dave Nagel played with it some in his review:

 

<http://www.creativemac.com/2002/09_sep/reviews/optipix102020912.htm>

 

Actually a better example comes in with a portrait where the skin tones are fairly light. If you only add in the dark edges (the places where the sharpening is negative) you enhance freckles, lines, etc. without causing highlights to bloom.

 

Another example would be a black lab. If the light edges are added to the dog, you can see detail in otherwise dark areas without causing funky shadows next to bright edges.

 

Regarding the blur in shadows, what if you put the luminance image into the quickmask, inverted the quickmask, and then blurred the image? This would weight the blur by the amount of selection (which was the luminance) and would selectively blur the dark areas more.

 

-Chris Russ

http://www.reindeergraphics.com