LinearSharpen 3.42

[Roger_Cavanagh]

Everyone and especially Pekka,

Thought I'd start a new topic as the old one's labelled 3.12 and I've come across at least one confused person. :)

Just so no-one has to go looking, here's the link to the 342 home page:

http://photography-on-the.net/D30/linear/

I have posted my assessment of 342. I'm not entirely happy with it, but I decided I'd better post something and fix it afterwards:

http://www.rogercavanagh.com/linear_comparisons2/

It doesn't reveal any secrets that we didn't know - Pekka has produced a stunning piece of work! :D But I do show 3 different examples (each in 2 sizes) showing alternative processing to LS. Viewing the versions side by side helps highlights the pros and cons.

I processed 25 images (not all posted in my review) every which way over the last few days. While it would be true to say that some 342 images would benefit from additional processing, any such processing would be in the nature of light polishing and certainly not major surgery. I will be very surprised if someone comes up with a major objection to 342.

How are you all getting along after a few days practice?

Regards,

[evan]

Roger, thanks for the review and comparison.

Pekka, thanks for the 3.42, great job !!!, the best so far (but i prefered a little bit more saturation)

Evan

[Pekka]

That's a well done review, thanks Roger. I was actually surprised to see how well 342 performed (I don't have e.g. LP to compare to).

If you see that you _always_ need a bit more saturation, you could easily add the step to the end of 'post'.

[John Boyes]

Thanks for the tests Roger, they certainly prove that 342 is the most accurate yet, and I thought 312 was neutral!!!

I have a number of images from portraits to product stuff that I've been using it on and everything looks just great.

I just need a new processor now, I'm sure my P3 450 is going to explode whilst batching in HQ mode!! :)

Cheers again for a fantastic action Pekka.

[Dale]

Thank you Roger for taking the time to do a review with samples and thank you Pekka for 3.42.
Dale

[John - NJ]

Some people have e-mailed me asking how much to send Pekka. Since Roger has compared Pekka's LS to Fred Miranda's LP, we can use that as a benchmark. Fred's LP action costs $20.

If LS was the only thing Pekka has contributed I'd say that $20 was about right. For me he has contributed much more. You decide what you feel is right.

I really think everyone could send something, no matter how small. The gesture alone will be quite gratifiying to Pekka, I'm sure. Please make this minor effort.

[ianmcg]

I agree with Roger that Aftershave 3 is very helpful with skin tones in particular. In fact I have yet to find an image that doesn't look better with AS3 applied. Even before I tried AS3 the skin tones in Roger comparison looked a bit off although I accpet the Q60 looked good.

For fun and by way of a contribution to Pekka's effort I have compressed the 5 "post" channel mixer action steps into one. Mathematically it can certainly be done the only issue is the effect of any rounding (which in 16 bit should be irrelevant) and the fact that the channel mixer only accepts integers compounding the rounding issue.

As it happens the channel mixer runs fast enough that it doesn't add much processor time to do it in five steps - but those of us who remember the early days of assembler programming know the joy of optimising every last bit of code.

I defy anybody to spot a visual difference between the one step and five step version.

Try Red = 74%r + 22%g +4%b
Green = -38%r + 126%g +12%b
Blue = -9%r + 16%g + 93%b


Just my 2c contribution. Otherwise thanks to Pekka again for the other 98c.

[Roger_Cavanagh]

Pekka wrote:
If you see that you _always_ need a bit more saturation, you could easily add the step to the end of 'post'.

Pekka,

Of course, that's true, sod's law being what it is that isn't the case. Otherwise, of course, you'd have built it into 343 (sic). :D

I was, by no means, systematic about testing changes in saturation, but when I did play around with the images, I have to say that Fred Miranda's Digital Velvia Pro did a really nice job of enhancing colour.

Regards,

[Sean Palmer]

Pekka,

Only looked in to see how everyone was getting on with 3.12, could not believe it when I saw 3.42 posted and also that Roger has managed to put an excellent coparison test together - sounds good. looking forward to trying this one now!

Many thanks to all,

Sean Palmer,

[Dale]

Roger_Cavanagh wrote:
I was, by no means, systematic about testing changes in saturation, but when I did play around with the images, I have to say that Fred Miranda's Digital Velvia Pro did a really nice job of enhancing colour.

Regards,


Roger at what stage did you run Velvia? Before converting to Adobe or 16 to 8 bit wth 3.42?

Thanks
Dale

BTW your web site contains some very helpful info. :)

[Roger_Cavanagh]

Dale wrote:
Roger at what stage did you run Velvia? Before converting to Adobe or 16 to 8 bit wth 3.42?

Dale,

I ran DV after conversion to Adobe RGB, which I did immediately after 342. I'm sure I must have had a reason to convert to Adobe RGB rather than leave images in Wide Gamut RGB, but it escapes me for the present. :)

BTW your web site contains some very helpful info. :)

Thank you very much.

Regards.

Roger

[bernardoadam]

Hi Roger,

I saw today your work. Thanks for it. I did a similar job last week with 200 fotos. My result was at the first view that Freds workflow is a little bit away in the color. Later I compared only Canon Software to the Pekka 3.12.

Here it sometimes a question of taste, some pictures I found better with Pekka WF and others I preferred the colors of the Original Canon.

I did a picture of my hometown where I definitely can say that Pekkas sharpening (I did High Sharp) did a little bit too much. Here I preferred the unsharp mask form PS.
At the end my conclusion is, that the best, but longest way, is, to process every picture individual. Lets say after a workflow of automatic conversion there has to be done an individual work on the valuable pictures.

Thanks again to bring your results to the web and also thanks to Pekka for his great job

Regards
Bernd

[Roger_Cavanagh]

Bernardo,

Hi!

bernardoadam wrote:
I saw today your work. Thanks for it.

You're welcome. :)

I did a similar job last week with 200 fotos. My result was at the first view that Freds workflow is a little bit away in the color. Later I compared only Canon Software to the Pekka 3.12.

3.42 is definitely better than 3.12

Here it sometimes a question of taste, some pictures I found better with Pekka WF and others I preferred the colors of the Original Canon.

If you mean that you liked the colours better, I agree with you, but the conversion should be trying to maintain colour veracity. Then we can play with warmth, saturation, etc., etc. At the moment, for me, 3.42 is the most likely to give the best result.

I did a picture of my hometown where I definitely can say that Pekkas sharpening (I did High Sharp) did a little bit too much. Here I preferred the unsharp mask form PS.

Have you tried the Chooser option. I have been playing with this on some favourite images from my archive. Processing them one at a time using Chooser and comparing the different levels of sharpening. My taste is for sharp, crisp images, so I look at the extra high first, then work down. Many images are over-sharpened with extra-high, but some are OK, and normal is invariably acceptable, even when high or extra-high look better (to me).

At the end my conclusion is, that the best, but longest way, is, to process every picture individual. Lets say after a workflow of automatic conversion there has to be done an individual work on the valuable pictures.

Ain't that the truth? :D

Regards,

[Pekka]

ianmcg wrote:
I agree with Roger that Aftershave 3 is very helpful with skin tones in particular. In fact I have yet to find an image that doesn't look better with AS3 applied. Even before I tried AS3 the skin tones in Roger comparison looked a bit off although I accpet the Q60 looked good.

For fun and by way of a contribution to Pekka's effort I have compressed the 5 "post" channel mixer action steps into one. Mathematically it can certainly be done the only issue is the effect of any rounding (which in 16 bit should be irrelevant) and the fact that the channel mixer only accepts integers compounding the rounding issue.

As it happens the channel mixer runs fast enough that it doesn't add much processor time to do it in five steps - but those of us who remember the early days of assembler programming know the joy of optimising every last bit of code.

I defy anybody to spot a visual difference between the one step and five step version.

Try Red = 74%r + 22%g +4%b
Green = -38%r + 126%g +12%b
Blue = -9%r + 16%g + 93%b


Just my 2c contribution. Otherwise thanks to Pekka again for the other 98c.


Hey that's interesting. I'm lousy in maths, so all help is welcome! :) What's the algorithm, is can't be plain mean values or those five R's G's and B's, can it?

What has always made me mad in all dialogs in PS is that they are not related to bit depth at all. The levels and curves should be much more accurate with better user interfaces and channel mixer should have decimals, and histogram should be nonmodal and realtime...

I'll have to try your mixer after I get this flu fixed. Thanks!

Pekka

[ianmcg]

It can be done because the channel mixer is only every linear combination and a multple st of linear combinations is still a linear combination.

Basically it works because the channel mixer is just a matrix multiplication between the 3x3 matrix of settings (lets call it M) and the colour values at a given point (lets call them V).

So the ouput is MxV where M is 3x3 matrix and V is a 1x3 vector. If you don't now how to actually do the multiplication don't worry about it.

The point is when you do a chain of channel mixer commands your effectively doing the following from the inside out:

M3x(M2X(M1xV))) etc. for as many Ms as you like.

Its a fact of matrix arithmetic that the above is equal to

(M3x(M2xM1))xV

Now since all the Ms are constant (i.e. your mixer settings) we can calculate one big M in advance which has exactly the same effect as the whole chain of Ms. Which I did in Lotus and voila.

As I say there is a very small issue of rounding Photoshop since the channel mixer only takes whole numbers but I doubt the human eye notices an error of 0.5%.