Tutorial: Sketch, Miscellaneous Insights (Mark SWEngineer)

[SWEngineer]

Sketchy thoughts:

How does the standard (decolor), copy, invert, GaussianBlur in ColorDodge mode really work? The best explanation I've seen comes from this tutorial. I have an alternate perspective to share, one that I think opens up some new ideas for sketch effects. (My perspective arose via these excellent tutorials from byRo: tut1 tut2 ) Rather than think of GaussianBlur as dispersing pixels, think of it as subtracting the high frequency (fine detail) from the original image. The lower the radius, the less detail removed. With the details removed, they no longer "mask" (dodge to white) themselves in the layered image, allowing them to show through.

Thus you should be able to (and do) get a very similar result this way: Create a layer filled with white below the image. Set the image layer mode to LinearLight. Now run the Other->HighPass filter on the image layer and adjust the radius to get the result you want. in some ways, this may be a superior method. In my test example, image elements appear more uniformly with changing radius. (I speculate this arises from subtleties in the math / algorithms in the Photoshop engine.) Try it yourself. The example image comes from this thread and has a nice mix of "frequencies" to illustrate my point. (I hope sdubose99 doesn't mind my using it here.)

Now that we recognize the invert / color dodge trick as simply extracting the high frequency information, we can bend it more to our will. For example, if you have a noisy or very busy picture, you likely don't really want the highest frequency elements captured in the "sketch". To prevent this, first apply a suitable Gaussian Blur to your image before doing the sketch treatment. (you can use Median, SmartBlur, etc, instead but the mathematical yin-yang relationship that exists between Gaussian Blur and High Pass no longer holds up. Not that this matters, the goal is a pleasing image, not elegant math.)

We can now be selective about the "frequency" content we expose other useful sketch filters too. Such filters include PosterEdges, InkOutlines, SmartBlur-EdgeOnly, FindEges, TraceContour, GlowingEdges, Photocopy, & TornEdges (one of my favorites). In fact, you can combine different treatments on different portions of the frequency spectrum. To give you some ideas, see the attachments. Note that I did NOT try to pick useful frequency ranges, just an arbitrary selection of high, medium, & low. Judicious choice of frequencies and filter settings should give a wide variety of good outcomes. My examples, being arbitrary, likely represent not-so-good outcomes. But, I think they illustrate the point.

-Mark

Attachments:
1) Comparison of HighPass vs ColorDodge method.

2) TornEdges filter variations (Applied to the right frequency range & with careful selection of filter parameters, this gives a really good broken line, multi-shade of gray outline type sketch.)

3) Composite Sketches (using the color image - no desaturation)
Bottom Layer: GBlur 4, InkOutlines 26/0/21
Middle Layer: HighPass4, GBlur 2, TornEdges 24/11/10, Darken mode, 41% Opacity
Top Layer: HighPass 2, FindEdges, LinearLight, 19% Opacity

[byRo]

Very interesting, Mark, nice to see that the tutorials got you thinking!

Some comments:
- While it is true that tendency of the inverted Guassian blur layer is to produce a High-Pass-like result, the fact that the "normal" sketch method uses a blending method of "Color Dodge" breaks the yin-yan (liked that) relationship.

- If you followed the "normal" procedure but were to use "Linear Dodge" instead of the Color Dodge you'd end up with exactly the same result as in your High-Pass method. That being so if you set up the "normal" layers, then you can compare the two methods just by toggling between the blending modes.

- I think I've never used Color Dodge for anything else but for doing these sketches. It is a non-linear mode (i.e. I don't know how it works) that seems particularly suitable for this particular application and, up to now, I had never given any thought to using anything else. In other words, it is the way that the Color Dodge mode gives us an pleasant visual result that is the magic trick in the "normal" method and not the use of the inverted Guassian blur.

- The method you used, with a White layer beneath the High-pass layer in Linear Light mode, will result in a pale image which will have to be enhanced by a levels layer. As such, then you really don't need the White layer, just use a levels adjustment with a white point at 127 or a bit lower and you get the same effect.

- I played with this a bit and, as you said, being a more "open" method it is easier to invent new things (a lot simpler than inverted-guassian-blur-with-color-dodge).
One thing I noticed - seeing as how this method doesn´t seem produce many grays - you can just put in, above the High-pass layer, a threshold layer at around 125, quick and simple.

I think you've opened up an interesting subject here, and I know I've got some thinking to do now with the new possibilities that have opened up.

Thank you, Mark.

Ah, and by the way, the "Composite Sketch" is very good!!

byRo

[SWEngineer]

Thanks for the reply, Ro. I hope we can get some good discussion / input from others in this thread. I think you already understand most of my counterpoints, but others may not, so...

Originally Posted by byRo

If you ... were to use "Linear Dodge" instead of the Color Dodge you'd end up with exactly the same result as in your High-Pass method.

I did this as a "thought experiment", just assuming they would give equal results. Thanks for the validation.

Originally Posted by byRo

it is the way that the Color Dodge (CD) mode gives us a pleasant visual result that is the magic trick in the "normal" method and not the use of the inverted Guassian blur.

I thinks it's arguable how "pleasant" the visual result is from the CD method. As I showed in the 1st attachment, at a given radius, the CD mode gives much less uniform density to the sketch than HP. I personally found this 'not pleasant', but I'm sure it may be beneficial in some images or to some tastes. So for me, there's no magic in CD vs HP.

Originally Posted by byRo

The method you used, with a White layer beneath the High-pass layer in Linear Light mode, will result in a pale image which will have to be enhanced by a levels layer. As such, then you really don't need the White layer, just use a levels adjustment with a white point at 127 or a bit lower and you get the same effect. ... One thing I noticed - seeing as how this method doesn´t seem produce many grays - you can just put in, above the High-pass layer, a threshold layer at around 125, quick and simple

I don't see HP / LL being paler than the CD method. Nor do I think it gives fewer gray shades. (The non-uniform density issue I noted above may give this impression, however). While for some effects the levels or threshold layer you propose may be warranted (or may help show you what you've selected to work with), I wouldn't use either as a default; many of the 'artistic' filters work fine applied directly to a HP or HP/GBlur treated image. And, as you demonstrated in your tutorials, you can use these directly with various layer modes & opacities. The 'openess' of this approach was what I wanted to emphasize. Aside: if you want grayscale in your sketch lines, the TornEdges filter (with optimal settings) directly transforms the HP or HP/GBlur image wonderfully.

Originally Posted by byRo

I think you've opened up an interesting subject here, and I know I've got some thinking to do now with the new possibilities that have opened up.

I'm honored to have inspired you and eagerly await seeing what you do with this. But I think I am still the student and you the master. So, I hope this doesn't delay any new tutorial of your own you've been working on.

Thanks for the complement on the "Composite Images" picture. I did kind of like the melting look to the houses. I'm not sure I like the halos however.

-Mark

[Patricia]

Thanks for sharing your thoughts with us, both of you! Having a technical background, I find discussions like this fascinating (and they really enhance my understading of how PS works), but I just don't have the time to experiment right now.

Patricia

[byRo]

Originally Posted by SWEngineer

I thinks it's arguable how "pleasant" the visual result is from the CD method....

"Pleasant" in as much as it pleases the user. It's one of those evolution things - why should anybody use this half-forgotten blending mode when a simpler approach can give a similar result - it's just that in many (of course, not all) cases the "CD method" gives a better (as interpreted by the user) result.

Originally Posted by SWEngineer

I don't see HP / LL being paler than the CD method. ....

No. it may not be paler. What I was saying was that (in both methods) you will probably have to use a levels layer on top to get the result you want - in which case the White layer could be substituted by setting the white point of the levels layer below the half-way point.

Originally Posted by SWEngineer

Nor do I think it gives fewer gray shades. ....

If you set up the "CD method", change the blending to Linear Dodge and look at the histogram, you will indeed see that there are fewer levels in the result.

Originally Posted by SWEngineer

....many of the 'artistic' filters work fine applied directly to a HP or HP/GBlur treated image. And, as you demonstrated in your tutorials, you can use these directly with various layer modes & opacities. The 'openess' of this approach was what I wanted to emphasize.....

Exactly

Originally Posted by SWEngineer

..and you the master......

Think I came up short here - a master would know what "Torn Edges " is, and I don't!!

Cheers,
Rô

[SWEngineer]

Sorry for being sloppy in my discussion, but typing the menu path to everything gets tiresome. This filter is found through these menus: Filter>Sketch>Torn Edges ... (I'm using Elements2, I expect a not too old version of Photoshop would have a similar path.) I've found this filter quite useful, at least following highpass / bandpass filtering.

-Mark