Tuesday, August 4, 2015

Build 269-A final, for a couple months, wavelet tool

In earlier tutorials I've worked with raw files I've taken. Now, for several reasons including some consistent criticism that I haven't written the book that explains the intricacies of every slider and curve in tools as complicated as RT's wavelets, my tutorials will use raws and jpgs available on the internet. Then anyone can duplicate my setting and if they are interested in the other settings...go for it. Just come back and tell us if you discover something interesting.

My source is http://www.photographyblog.com/reviews/   They have a very large collection of raws, often with matching jpgs, from a multitude of cameras. The images I used for this tutorial shows a family group at the end of a subway corridor that are almost lost in the darkness and ISO128000 noise. A challenging shot that will showcase several RT tools.

The link for the jpg--we will compare the RT results with the straight out of the camera results--is
http://img.photographyblog.com/reviews/panasonic_lumix_dmc_g7/sample_images/panasonic_lumix_dmc_g7_36.jpg  The matching raw file is panasonic_lumix_dmc_g7_36.rw2

If you haven't upgraded to the windows build RT 4.2.269 or mac RT 4.2.270 do so immediately. If you are working with wavelets any earlier build will disappoint. Build 269 is stable (no crashes in about a week) and has several new features and bug fixes (a beta release for RT 4.3 maybe) plus some  new documentation.

Jacques, the chief developer behind wavelets and CIECAM has posted an updated RawPedia article in french. I'm sure is makes great reading if you are fluent in technical french, but after going through google translates  the meaning has been mangled in some sections. Example, when you come to a confusing sentence about a 'beach'  the correct translation is 'range' as in range of values. But for the most part it is  understandable and in this tutorial I will try to expand on what you will learn there

Wavelets are 3 dimensional. This mathematical third dimension is one of the reasons you see three sliders in many sub tools; two of which will confuse you because they seem to do nothing or make the image worse,  Some reasons for this behavior are:
1- they work much better with a different type of image
2-they are in the tool because Jacques and team have future surprises planned
3-they are  there because Jacques and team hopes we users will come up with some of the new surprises.

So if you hit a slider or curve that confuses you leave it at the defaults. They have been carefully chosen to work with the majority of images.

Enough talk, the workflow: Lets see what we can do with this  image

This is the matching jpg to the raw file we will use,  ISO 12800 and SOOTC, The goal is to create as decent an image as possible of the family group that are barely visible at the end of the subway tunnel without completely destroying the detail by the handrail  in the foreground


I started with the the residual image tool. It got its name because it was what was left over after the calculations extracted the wavelets. What you do to the wavelets doesn't change it. And vice versa. You must recombine them either on your RT screen in real time in a 100% view or after a trip to the queue. That's one disadvantage when developing wavelet workflows. To chase down all the smal artifacts I develop test jpgs at key steps so I can look at them with other programs. Such as imageJ, a widely used and free NIH scientific image analysis program.

You can modify the residual image with non wavelet tools.  The shadow and highlight sliders lighten or darken their areas of the images. Their matching threshold sliders decides how large an area they effect. In past tutorials I recommenced  using CIECAM for contrast and chroma because of the tool's curves. Now with the new final tuneup curves ...you pays your  penny and takes your pick.

With this image  the shadow slider brightens the family group so I can see how bad the noise is.


It is ISO12800 bad, clearly what Jacques calls high noise in his rawPedia article. I'm pleased that the noise reduction workflow I touched on in my Lorde tutorial  is now his recommended workflow. In the program flow the preliminary noise reduction in the second tab is now before the color conversion step. The wavelet denoise is now at the end and is intended to clean up any additional noise and artifacts created by sharpening and other adjustments.


For the noise preclean I chose Lum 70 and Detail 30. Depending on the final results I can fine tune these numbers.


To lighten the back area more I tone mapped the image. While wavelet tone mapping is less aggressive and more artifact free than RT's earlier version when it is combined with its gamma slider it can lighten and darken to both extremes.

It's worth mentioning that tone mapping works on the signal side of the all important signal to noise ratio just as noise reduction works on the noise side.  RT tone mapping uncovers noise in dark areas. It doesn't create it.  We are prone to saying the new full frame camera we bough to collect more light the light in its bigger pixels makes less noisy images,  Instead, when the exposure collects four times the light-- four times greater signal-- that is combined with twice the noise to double the signal to noise ratio. This may sound like super geek nit picking but I've seen forum discussions go off the rails into making wrong claims when people didn't understand or take into account the difference. A recent claim I've seen that tone curve also causes noise is one example.


To bring back the sharpness lost with the initial noise reductions I used edge sharpening with edge detection checked. That sub tool  tames edge sharpening. The first slider, gradiant sensitivity, determines the amount of taming,  The second noise slider checks for nearby noise.  With low ISO images it has no effect; with high ISO images it increases the sharpening and can create artifacts. The last slider depends on the image and the other settings and can increase or decrease sharpness.

I followed my own advice and used the default values in the subtool. I set the sharpness, changed first level to unchanged and dragged out another contrast curve before I pulled up a 300 % view for a good look at the increased noise.

If it had been much worst I might have increased the Lum settings. Instead I went to wavelet denoise


A combination of denoise and impulse noise reduction from the second detail tab--combining the two is a class 3 tip --cleaned up real noise exposed by sharpening and a few L shaped artifacts visible in 300% view.  Since the goal of this workflow was to fix the darkest area in a ISO12800 image we have to live with some loss of sharpening.

I left the bottom denoise slider at zero. Ever since I began working with wavelet I wondered why anyone would add a slider that increased the noise to a noise reduction algorithm. Jacques latest RawPedia article explained all.  Take a 30sec dark night image and the slider brings out the stars in all their glory. Another example of what seems useless or worse with one type of images my be very useful with another quite different type.

The developers didn't include all the sliders in a big joke to confuse us users. They all have a purpose and it's our job to work out the neat little photo tricks they allow.

In his RawPedia article Jacques also cautions about not expecting decent results if you use a saved pp3 on a different image. On this I think he is being over cautious. If you use a saved pp3 that was perfect for a close up of a rose in bright sunlight on one of the ninety concert hall stage shots of your favorite pop-rock band you won't be happy. But if you develop one of the stage shots and use it in the file browser to develop the other 89 stages shots you are well on the way towards fine tuning the three perfect shots you will post and brag about on your Facebook page.


I used a parameter curve in the new Fine Tune tool to even out the lighting in the corridor, Now the histogram is bunched up in the center and needs to be expanded for a better black to white contrast range. So I went to CIECAM.


CIECAM contrast and a touch of colorfullness fixed the problem, So how did we do?


Photographybogs.com's untouched camera jpg vs RT's greatly improved jpg.


A roughly neutral grey S/N improvement from about 10 to almost a 100...


at the cost of a very slight loss of contrast--see the lines on the meter--that no one would notice if the two images weren't shown side by side.


A scaled down version (for the blog) of the original photographyblog jpg and the RT jpg. If you aren't impressed you must be working for an enemy big bucks image editor company. (grin)

Thursday, June 25, 2015

Wavelets and Lorde

After a major professional photoshoot--the cover and feature article in Australian Vogue--Lorde, the New Zealand songwriter and singer of Royals and Pure Heroine fame--posted a selfie on Instagram. So far 148 thousand of her 2.5 million followers have loved it

My 12 year old granddaughter didn't love it. Too, too noisy. So fix it. grandpa.  (Got her RT trained since she was three and I gave her a big girl camera that took pictures of her feet and her cat and herself in a mirror)

Grandpa's two minute fix. Far less noise coupled with slightly more sharpening. The S/N was measured using ImageJ


The idea is to remove as much noise as possible and not worry about the sharpness. So there is nothing magic about the  91.59 luminescence. Could have been 100% if this wasn't a two minute fix and I had been paying more attention to the GUI fit on the screen.


Increasing the edge sharpness strength to 34 restored the original sharpness. Some small artifacts showed up as bright spots at 300% magnification. Using the denoise section of the tool removed most of them. I should note that the  tool is still being debugged and improved so these minor artifacts may not be a problem in the next build.

Final note. The denoise slider numbers in the history panel are the correct numbers. The slider settings  on the right  are the defaults when I set the sharpening strength.

To download the latest version of RawTherapee go to http://rawtherapee.com/downloads
 It's is open source, works great and will be forever free

Monday, May 11, 2015

Wavelets A Third Look

I suspect that by the time you read this blog post it will be out of date. Things are moving fast in RT wavelet land and wavelet tone mapping is galloping up behind me.

In my last post I talked about a three tool workflow using only wavelets, noise reduction and CIECAM. In this post I'm stripping down farther and using only wavelets.  Along the way I'll add my opinion about the multitude of adjustments that may panic a novice user into slider shock.

First question. "Are wavelets worth the learning curve?"  Yup and double yup.  With only a few adjustments you can do this.

If you aren't already aware how blogspot works click on the following images to bring up a menu at the bottom. Original size enlarges the screen shot into something more readable.

In the wavelet setting section the strength slider combines the wavelet and the original image.  I haven't found much use for the combined images but moving the slider back and forth between  0 and 100 is a quick way to see how wavelets have improved (or degraded) your image.

You can create a stack of up to 9 wavelets in any image larger than 1024 pixels. Unfortunately this includes 100% detail windows. They need to be large when I'm working on a D7000 image.  Since my 24 inch 1920 by 1080 pixel monitor doesn't have that much real estate I zoom to 100% and forget detail windows.  But if I spring for a dual monitor card for my birthday--I already have a 1024 oixel 19 inch monitor--I might rethink that strategy.

Change file size to tiled if you are using a 32 bit machine. For a 64 bit machine with a decent amount of memory full image eliminates any tile artifacts.

The D in edge quality stands for the Daubechies function. It's a step up from the Mexican Hat function used by Gimp since you can change the number of coefficients.  It that mathematical  talk leaves you confused the thing to remember is that you don't notice changes until you blow up your image to about 400%.  So if you aren't planning to use a very small crop or print a very large poster leave it at the default  setting. (See Update below. RT keeps improving.)


After playing around with wavelet I decided my Nikon D7000 and its kit len are a bit soft. Seven or eight clicks on the contrast + button improved things. I've also changed the default 'apply to' box from adding contrast to only the shadows and highlights to adding it to the full image. Since this improved all the D7000 images I have tested so far I'll include this as a default when I create my custom three tool profile. So use wavelets to check out your camera and lens combination to see how they perform.

If you are doing conventional development don't bother opening the preview image section. But if you occasionally slip into the Crazy Artist mode it's a double WOW place to start. I'll give an example at the bottom of this post.


There are three chromatic sub tools.  The pastel-saturated tool links the chroma changes to the contrast settings I've already set.  The slider shifts the effect from pastel tones to saturated tones. It works but has a limited range. In my first wavelet post I discussed both this and the nine point 'curve' that was replaced by the sliders in the next sub tool.

With  this sub tool  you can set the chroma change for each wavelet. I suspect if  I was working on another image with a range of details that  I wanted to make more colorful and another range of details that I wanted to mute this would be a useful tool. But don't expect big changes.

For landscape images I use the third sub tool.

The single slider goes from no changes to oversaturated. This will be the default in my three tool workflow.


Edge sharpening is much improved.  In my first post I dismissed  it with "a little goes a long way." It wasn't finished and created an artifact lover's heaven, This image shows the changes I've made so far without wavelet sharpening.


With 30% sharpening   All edges are sharper. But we are beginning to add sharpening noise to the walkway in front of the building. Wavelets haven't magically ended the battle between 'really-sharp' and 'lot-of-noise'.

Here is the noise at 65%, Sharp but noisy. At 100% it is much worse.


My compromise in the battle was a sharpening of 43%,. Note how the blocks of the wall now jump out of the image and the sidewalk is reasonably noise free. To clean up that noise I used the denoise/ refine tool.  Unfortunately I was interrupted and forgot to save a screenshot.  The tool is simple, just a check box and three of the new combined sliders that control different denoising algorithms. The top setting is strength. The bottom is detail.

With this image level 2 worked best over all. But to remove all the noise residual in the blue sky I needed a small amount of level 1.

Leave the edge detection box unchecked.  The way it is setup now it blurs the image. This maybe because the defaults are wrong. Or it may need more work. It will, I suspect, be used to fine tune the noise and edge detection

{Update. With RT 4.2.173 this sub tool works far better--or I'm better at using it on an image that has a lot of detail. The default strength should be 0,  not 82, as I suspected.  The noise threshold slider cuts noise when used with the noise/refine tools.  The threshold high slider sharpens edges, mainly by increasing the local contrast. And it now has a visible effect on 100% images. especially when edge quality is set at D14 high.}

To finish I straightened the image to eliminate the jags on the verticals of the doors and windows


As a comparison I moved to the top and brought the strength slider down to 0.  Like I said earlier wavelet sharpening is very much improved.

I'm skipping over the gamut tool. It's supposed to have a much bigger effect on wavelet tone mapping so I save it for another post. And to be honest I haven't worked out how or why I would want to use it.


The residual image is, to quote, "what is left over once you extract the wavelets."  In practice that means nothing you do to the wavelets will change the residual and vice verse.  To modify the residual the wavelet tab come with a set of non wavelet tools.

As expected the first four sliders lighten or darken the shadows and highlights with the threshold sliders determining what tones are considered shadow and highlights. The contrast and chromaticity sliders also work as expected and the HH curves, set for blue sky, turn it greenish or purplish when I drag the blue control point up and down

When wavelets are finished I may discover a big reason for using this tool. But if I hadn't decided to everything with wavelet I 'd be using CIECAM for these adjustments.  The extra sliders and curves gives me much more control over the look of the final image.


I'm not going to attempt to illustrate  what you will see when you play around with preview setting. At the default--all wavelets viewed in all directions-- you see your image.  With the other setting you can see each individual wavelet or the stack of wavelets above and below that wavelet. Then you can scan the stack vertically, horizontally, diagonally or in all directions. Then you can set three different background: the unmodified residual image, the grayed out version and the blackened version.  Add the fact that what you see depends on such things as the local contrast of your original image and you have some idea of the number of combinations.

This is a stack, level 2 and I,  viewed vertically against a grayed residual image.


And this is the same thing viewed vertically.  A bit of a difference.

Some time after midnight as I was about close up, my Crazy Artist muse conked me on the head and yelled in my ear. "Forget this one tool workflow crap. artist boy. Open up all them curves and sliders and buttons. Move them control points around. Go for the ugly. Go for the weird. Go for the crazy crazy, crazy til you got something so crazy I'll let you to put my name on it. And if you don't get it exactly right first time, don't worry. I'll be back to make you do it again. Cause I'm an addictive muse, artist boy."

Best viewed original size

While this post was written using RT 4.2.151 get RT 4.2.173 from the download page
There's been some fixes and improvement but the layout is unchanged.


Monday, March 9, 2015

Second View Wavelets--The Workflow

Yesterday in my last tutorial I outlined a possible default workflow based on wavelets. Today I experimented around to see how well it works.

Photo loaded

This image  was taken at in 9-11 memorial that was traveling the county by truck last summer. I shot it handheld with my D7000 at ISO6400  1/100 sec.


Around 6 clicks on  the contrast + button. The D7000 produces soft raw file straight out of the camera

Fifty Percent Wavelet Sharpening

After a certain point more sharpening only increases noise and artifacts. With my camera I've been switching between 25 and 50 percent. At some point I'll settle on a default setting.

Noise Reduction without Median Filter

A decent looking result until you increase the magnification The ratios for the luminous noise sliders were 80 strength, 20 detail which is higher than my usual 50-50 ratio,


Scattered line artifacts are more obvious at 200% magnification especially if you blowup the screen shot.

3X3 Strong Median Filter

The median filter does a good job reducing the artifacts. There is a small lose in detail but the overall wavelet line sharpening is not affected much.  I'm not ready to scream from the rooftops-just from this blog post- but I beginning to think you don't have to trade detail for low noise anymore in high ISO images 

A BIG BIG advance for low-light no-flash handheld photographers like me. Perhaps one of the developers could comment on this dreamy idea.

CIECAM  color tune up

I used the curve and lightness slider to darken the area around the poster. I used the contrast and chroma sliders to bring up the colors on the woman's blouse.

The final edit vs the original

Bottom line.  My D7000 raw files need more contrast and sharpening when they come out of the camera.  My high ISO raw files also need noise reduction. So these three operations can be combined into default profiles similar to the three default versions that come with RT

CIECAM tuning and cropping are too image dependent. But the GUI could be set up with these tools ready to go and  everything else  closed up. With that setup I'd be willing to bet that I could process 90% of my images with just these four tools, two of which work automatically.  

First View--The Wavelet Tool

[EDIT  (June 30, 2015)  RawTherapee's wavelet tool is moving ahead at warp speed.  An example. Two day's ago I asked a developer about sharpening artifacts and wondered if they might disappear in a future build. A couple hours later I got a message that a new build, RT 4.2.231, was up on the RT download page.

The great news was that this build fixed the artifact problems.  The bad news is that I was in the middle of researching a fifth look at wavelets and have to redo my research.

And no, I don't have a private in with a superfast developer. He had been working on the problems for a while.  And yes there are three more tutorials after this one with a fifth being written.  The second is on a potential default workflow; the third is on developing a well exposed landscape shot using only wavelets; and the fourth on cleaning out the noise in pop star Lorde's latest widely shared selfie.

The tutorials have become more dated as the wavelet tool evolved so the screen shots aren't  accurate.
And in the three day since I posted the Lorde tutorial I've discovered a better way of cleaning up high ISO images using the new build--one I will explain in the next tutorial.

So upgrade to RT's latest build so we can continue to explore the exciting landscape of wavelet land.]

My book on wavelets with the equations in an appendix--one written for non math types like myself--only runs 250 odd pages. So I may eventually slog through the non math pages. Wavelets are complex and don't lend themselves to short summaries. Still the one big thing I've picked up so far is that wavelets are transforms.

Just as  jpgs are transforms  When you convert a raw file into a jpg  in the first half of the process the algorithm  creates an intermediate state, the color information. If  RT's software allowed you to do a reverse transform -the second half of the process- without changing anything you would end up with the same size image you started  with.

Something not very useful. So you have two sliders to changing things. If you are like me, you leave the subsampling slider on balance and forget about it. The second slider, quality factor, sets the size of the jpg by finding pixels with colors that are 'close enough' and coding them into a single value. If you go overboard and set the quality factor too low for very small jpgs 'close enough'  becomes "way too much" causing banding and other artifacts,

With  RT's wavelet tool, instead of colors you are dealing with spacial frequencies. Think Contrast by Details  with much more powerful sliders. With them you can fine tune your raw or your jpg and approach that ultimate goal- the perfect image. Or if you flip into Crazy Artist mode you can create some insanely weird images .

I created the images for this tutorial using developer's build 104. As might be expected, not everything worked.  Big things like CIECAM which crashed RT. Or smaller things like edge sharpening and color adjustments that need more work. Plus there is a warning that the pp3  format will be changing and any wavelet editing done with this build won't work with the final build. Despite all this, the build was worth studying and blogging about.

I was about to post when developer's build 113 appeared in the forum. It has a set of neat fixes. Like a working CIECAM  and improved edge sharpening. So instead of following a single image from loading to final development I'll be sticking in additional images highlighting what I discovered

In this screenshot I've set the strength at 100 percent--only wavelets.  Strength is a useful slider since if you go a bit too far with the wavelet adjustments mixing wavelets and your original image can help cleans things up.

Of the 9 possible wavelet levels I'm using the default stack: 7 wavelets starting at level 1, the finest detail, and going on up to level 7, the coarsest detail. Levels 8 and 9 where the 'detail' is essentially the whole image controls chroma when colors are linked to these sliders (see below).

If you have an older 32 bit system you can use tile size to reduce the memory needed.

You are looking at level 1 through 4 scanned in all directions. I could have scanned horizontally, vertically or diagonally to create different looking images. While CrazyArtist images belong is different tutorials this is a good time to say that starting with a stack of wavelets can lead to some unusual images.

Wavelet image editing is not new. Google Gimp and you can find several longish tutorials on how to remove skin imperfections by modifying and combining layers. While this approach is not insanely difficult, RT is the only editing system  I know of where you have everything you need in simple tool.

The jpg show a  model with excellent complexion (or one who has been well photoshopped) but nobody is totally perfect. After going back to "all levels in all direction", the setting for working on the complete image, I increased the contrast of the first 4 levels. This brought out a mole, (the arrow)

Bringing the sliders for level 3 and 4 (the spacial frequency of the mole) into the negative range eliminated it.

With the 104 build  when looking at the spot where the mole had been with high magnification you could see an artifact that didn't match the surrounding skin tone. With the 113 build the match is much smoother. Another fix.

You can change the chroma with this curve.  Your options include 'all chroma' and 'pastel- saturated chroma'.  The second option adds a slider that direct the changes from the pastel to the highly saturated colors. The third option, a non curve mode, linked chroma to levels 8 and 9.  This makes far more drastic changes.

I've dragged the curve up to increase the chroma.  Dragging it down would have decreased it.

This curve will be reworked.  In the final release it with be replaced with another type of curve or with a set of sliders. These will allow you to set the chroma of each wavelet level separately. Something I suspect that will make some interesting looks.

Edge sharpening is another use of wavelets. With build 104 a little went a long way. The image started to create artifacts at about 8 so using it didn't sharpen much.

 A big fix showed up in build 113, Now you can do this.

No sharpening vs. 100 % sharpening.

Coupled with NR and its median filter wavelet sharpening now works over its whole range. With not a trace of a halo. Good by unsharp mask.

At the bottom we have the gamut controls and residual image or color information. Here you can lighten or darken  the shadows and highlights and control contrast and overall chromaticity. These sliders work as expected.

With build 113 for the fine tune you can leave the bottom two sliders at zero and switch to CIECAM . Its numerous curves and sliders gives you far better control over the look of the image.

The toning section in both builds hasn't been implemented. These color tab adjustments worked as expected.

On the other hand the vibrance tool modifies wavelet edited image more than ordinarily edited image.

In the end I finished with a black and white edit

Original image joined to the wavelet processed image.

Where are we at now.  Once the wavelet tool is finished, I suspect it will become the core of my workflow. This shot of girl-scouts during butterfly month at the local gardens was shot at ISO 3200.
The editing steps were:
1-auto exposure
2-about 6 clicks on contrast +
3-sharpening and noise reduction
4-switched to CIECAM to correct colors using a brightness parameter curve
5-final slider adjustments
7-using this image's profile to batch process a set of similar images.

Bottom line,  A short, quick and high quality workflow. Wavelets RULE!

If you want to play around  you can find developer's builds in the RT forum.  Just remember these edits aren't going to work with the final version.