The effects of this tool are visible at any zoom level. However, due to the nature of the algorithm, only the 1:1 (or more) preview zoom will match the saved image perfectly. The size of the image has a direct impact on the perceived sharpness. In practice, this means that you should use this tool at a zoom level similar to that of the desired final size of the image in terms of viewing distance, screen or paper size, etc. So if you intend to print a high resolution image on a 90x60cm canvas and view it from a distance of 30cm, then it is reasonable to zoom in to 100% and adjust the "0 (Finest)" contrast slider. However, in reality such large prints when hung on the wall are observed from a good distance off - say, from a couch some meters away - and at this distance adjustments to the finest detail level cannot be distinguished; the eyes cannot perceive such fine details at this distance. The same applies if your image will subsequently be resized, e.g. for email attachments to friends or customers, for image sharing websites, for social platforms, etc. Not only will the image resolutions be lowered but the images may be viewed on devices with a lower resolution, not even fullscreen, on laptops, tablets or smartphones. Keep this in mind, so you do not waste time adjusting detail levels which will be lost anyhow. Usually only Contrast sliders 2 and above provide a useful effect.
The wavelet tool is close to being finished (today being 2015-02-11). The information contained herein should be mostly accurate for the wavelet tool as available in version 4.2.217. Some information still needs to be written or updated.
- 1 What is a Wavelet?
- 2 Utilities
- 3 Show wavelet levels
- 4 Contrast
- 5 Chromaticity
- 6 Split Toning
- 7 Edge Sharpness
- 8 Gamut - Controls
- 9 Residual Image
1 What is a Wavelet?
A wavelet, or more specifically a wavelet transform, is a mathematical function very useful in image processing as it enables one to split the signal into various levels of detail so that you may work on the levels which are of interest to you. The English term "wavelet" was introduced in the early 1980s by French physicists Jean Morlet and Alex Grossman. They used the French word "ondelette", which means "small wave". Afterwards, this word was transferred to English by translating "onde" into "wave" giving "wavelet". The wavelet transform is fairly similar to a Fourier transform, the general difference being that wavelets are localized in both time and frequency whereas the standard Fourier transform is only localized in frequency.
Decomposition in RawTherapee currently uses the L*a*b* space. This decomposition produces two kinds of fundamentally different products:
- Several levels of detail. Choosing the number of levels depends on your needs. The first level corresponds to an area of 2x2 pixels, the tenth level corresponds to a region of 1024x1024 pixels. The more levels, the more processing time and memory is required. Each level contains only the "variations" (positive or negative, for example the contrast that gives data type 0+D1, 0-D2). If a picture is absolutely uniform in luminance and color, each level will contain no information. Thus, one can act separately on each of the levels, but with specific algorithms that take this feature into account. The "quality" of information present in each level is a function of the mother wavelet used: either it is simple as is the case with the "Mexican hat wavelet" as used in dcraw and GIMP, or it is more complex and powerful as is the case with the "Daubechies wavelet" as used in RawTherapee.
- A residual image. This image corresponds to the difference between the source image and the combined details of all level scales. The residual image has the same characteristics - in terms of its treatment - as the source image, and lacks the details which are present in the other levels. Modifying the characteristics (contrast, chroma, etc.) of a level has no effect on the residual image, and vice versa.
Depending on our needs, we will need to look either at an individual level of detail, at several levels of detail, at the residual image, or at all of these at once. RawTherapee makes this so easy it's ridiculous.
After decomposition it is possible to use wavelets for different purposes:
- Image compression,
- Noise processing,
- Secret watermarking,
- Specific treatment for astronomy, separating the background (residual image) from the planets and stars, or separating stars from dust clouds.
Wavelet decomposition is used to separate each luminance level (contrast), the chromaticity of each of the a* and b* channels, and the residual image. This action enables you to apply different treatments to the luminance detail levels, the chrominance detail levels, and the residual image which depends to the selected number of luminance detail levels. It is possible (for software developers) to extend the existing processes to others. For example, each level corresponds in fact to the equivalent of a layer; by having suitable GUI tools, it is possible to edit the details (stains, spots, etc.) only in the levels they appear in, and therefore preserve better image quality. Similarly, it is theoretically possible to apply treatments available in other RawTherapee tools to the residual image (L*a*b* Adjustments, Color Toning, etc.). Their implementation may require increased memory requirements and processing time.
You can change the type of mother wavelet used by going to Preferences > Performance & Quality:
- Check the option "Use Wavelet Daubechies D6 instead of D4" to increase the number of coefficients of the wavelet.
- This action has no or little effect on the residual image or high levels (5..10)
- ((can't make sense of this)) On the other hand it (it what? it=D4 or it=D6?) can reduce unnecessary stress effects ((what is this?)) for low detail levels (0..1) under undesirable areas that D6 is more discriminating than D4 (skies, high values of initial contrast, etc.), for example when using high values of "Edge Sharpness".
- There is virtually no difference in processing time or memory requirements between D4 and D6.
Once the "treatments" have been applied (to the levels and residual image), the image is re-composed.
RawTherapee has a tool called Contrast by Detail Levels - it "looks" like this Wavelet Levels tool but has several differences:
- CbDL has fewer levels,
- ((did I get this right?))CbDL lets you only work on the levels' luminance, while Wavelet Levels lets you also modify the chromaticity,
- CbDL has no residual layer.
Having said that, there is nothing preventing you from using both Contrast by Detail Levels and Wavelet Levels together. CbDL is located upstream in the toolchain pipeline, ((what does this mean?))and can interfere on levels 0-5.
2.1 Total Levels
This slider allows you to select the number of detail levels the image should be decomposed to. The greater the number, the more processing time and memory will be required.
Choosing a low number of levels, 5 for example, will limit the system to details of a scale of 32 pixels. The residual image will contain all the detail the image has except for that contained in levels 0-5.
Choosing a high number of levels, 9 for example (the 10th level is called "Extra"), will allow you to change the details of a scale of 512 pixels (level 8) or 1024 pixels (level "extra", whose effects are only visible if your preview size is large enough). The residual image will differ significantly from the original image as levels 0-10 contain all of the details, so what is left in the residual image is a blurry mask.
The residual image is what is left after all the detail levels have been removed, as many as you have chosen to use via the Max Levels slider. For example if you decompose the image into 7 detail levels (set Max Levels to 7), but you only use the contrast slider on the first three levels (on levels 0-2), the residual image is still whatever is left after the details of levels 0-6 have been removed from it, not just levels 0-2.
Important! RawTherapee uses as many levels as it can unless the size of the level is larger than the size of the image. That is to say, if you are working on a photo larger than 1024x1024px, then changes made even to the 1024px level will be visible in the saved image. If the photo is smaller than that, then the 1024px level will be skipped. This is important to keep in mind, as the same principle applies to the preview: if the photo is larger than 1024x1024px but the preview area is smaller than 1024x1024px, then the 1024px level will be skipped in the preview, though it will be used when saving the full image. This may lead to significant differences between the preview and saved image, so it is worth committing this caveat to memory. You may expand the size of the preview area by shrinking or hiding the left and right panels (shortcut "m") and you gain a few more pixels by making the RawTherapee window fullscreen (shortcut "F11"). The interface informs you how many levels are used in the preview under the contrast sliders, "Preview maximum possible levels=9" for instance.
2.2 Full Image vs Tiles
A selection box allows you to choose between:
- Full image,
- Big tiles,
- Little tiles.
It is always preferable to use "full image" as it avoids quality issues in the transition zones between tiles. If you do not have enough RAM or/because you are processing huge images then tiles may need to be used.
2.2.1 Memory requirements
Memory requirements, in bytes, for "full image" mode, assuming 9 levels are used, with examples given for two cameras:
- Pentax K10D (10.2MP) 3888 x 2608
- Nikon D810 (36.3MP) 7360 × 4912
- Minimal requirement for opening an image in RawTherapee, when all tools are off/neutral
- K10D: 116MiB
- D810: 414MiB
- When using level contrast, chromaticity or hue targetting/protection
levels*width*height*3 + width*height*7
- K10D: 329MiB
- D810: 1172MiB
- Additionally when using Avoid Color Shift
- K10D: 39MiB
- D810: 138MiB
- K10D: 483MiB
- D810: 1724MiB
3 Show wavelet levels
This panel has three combobox menus which allow you to refine what you see in the main preview. It serves to teach you, to help you understand how to work with the wavelet system. Changes made to this panel affect only the main preview, not the processed image.
The first combobox lets you choose to preview:
- One wavelet level,
- Below or equal to the level,
- Above the level + residual,
- All levels & directions.
The second combobox lets you specify the reference level for the above options, from "level 0" to "level 8".
The third combobox lets you choose to see the following wavelet directions for the first three preview options:
For example you can choose 'one wavelet level' mode to find the wavelet which contains the scale of details you want to work on (e.g. the level containing the scale of skin blemishes but not skin texture) and to see the effects of your adjustments on that level.
Another example, you could choose to preview "above the level + residual" with "level 8" selected to see the residual image and the influence upon it of the tweaks you make in the "Residual image" panel, explained further on.
4.1 Contrast Levels
By default, seven levels appear in this menu (starting from level 0). You can reduce or increase this number by changing the Total Levels slider.
Reminder that the fewer levels you use, the less RAM and processing time is required, and the more the residual image matches the source image because fewer detail levels are removed from it.
The "Contrast -" and "Contrast +" buttons provide an easy way of tapering off the level values in a gentle slop. Of course, these buttons aid in adjustments where small levels are most significant, but often one may want to adjust the high levels too, for instance so that the small levels increase contrast while the large levels decrease it.
Each level corresponds to a scale of details. Action on a given level is only possible if that level contains contrasting areas to begin with. If a level is uniform in contrast, the slider will have no effect.
- Level 0: 2x2 pixels
- Level 1: 4x4 pixels
- Level 2: 8x8 pixels
- Level 3: 16x16 pixels
- Level 4: 32x32 pixels
- Level 5: 64x64 pixels
- Level 6: 128x128 pixels
- Level 7: 256x256 pixels
- Level 8: 512x512 pixels
- Extra: 1024x1024 pixels
Remember what you read in the "Total levels" section above: the preview area must be at least as big as the level in order for it to be visible in the preview.
The residual image is not listed above because it is not a level - it is whatever is left over from the original image after the detail levels are taken away from it.
Changes to a level can be applied in a straightforward manner, or only to highlights and shadows. The same is true for changes to chromaticity. Read on.
4.2 Apply to
This module allows you to focus the sliders' action on a luminance range, for example increasing the contrast of fine details for high luminance, and reducing the contrast of coarse details for low luminance.
4.2.1 All Luminance
If you select "All Luminance", each slider has the same action regardless of the luminance of the area.
If you select "Shadows/Highlights" more controls appear in order to customize the action:
- A luminance range for shadows,
- A slider which allows you to select how many levels should only have a darkening effect, always starting from "level 8", even when you use less levels than that.
- A luminance range for highlights,
- A slider which allows you to select how many levels should only have a lightening effect, always starting from "level 0".
For example if you only use 7 levels, which makes the highest level be "level 6" (the count starts from 0), and you want "level 6" to only have a darkening effect and levels 0 and 1 to have a lightening effect, then set the Shadow Levels slider to 3 and the Highlight Levels slider to 2.
Changes to levels which are not targetted by these two adjusters apply evenly to their whole luminance range.
The construction of this module is similar to that of the contrast module. You have the following choices:
- All chroma: in this case, the whole chromaticity range is affected by each level's change, independent of contrast levels.
- Pastel/Saturated: here you can choose two limiting ranges for pastels and saturated colors, independent of contrast levels.
- Link contrast levels: here chromaticity changes are directly related to those of contrast.
5.1 All Chroma
If you select "All Chroma", the whole chromaticity range of each affected wavelet level is modified, regardless of how saturated it is.
Chromaticity = sqrt(a*a + b*b) is a wide open, but I limited it to 0..140, which should satisfy almost all cases.
The chromaticity curve:
- The x-axis represents the wavelet levels, from the finest on the left end to the coarsest on the right end. Its effect is therefore not continuous, but discrete: the vertical values are sampled in 9 points equally spaced out along the x-axis, starting at the very left edge of the curve, and ending at the very right edge. If you use less than 9 wavelet levels, say 7 points, the rightmost 2 sampled points would be ignored.
- The y-axis represents the force of the action for each level. If the curve is above the equator, chromaticity for the affected wavelet levels is increased. If below - decreased.
The same remark applies here as that for the contrast levels: action on a given level is only possible if that level contains varying chromaticity to begin with. If a level's chromaticity is uniform (does not vary), the curve will have no effect for that level.
You can treat the chromaticity in three ways:
- All chroma - Evenly apply the change in chromaticity (y-axis) regardless of what chromaticity the levels already have.
- Pastel/Saturated - Apply the change in chromaticity with respect to the pastel/saturated threshold value, default at 5, so the limiting range of saturated tones will be effective for the first 5 levels and the limited range of pastel tones for the remaining levels.
- Link contrast levels - Here the action on the chromaticity - insofar as it varies in the original image - will be directly proportional to the slider settings for contrast levels. This proportionality is achieved using the "chroma link" slider: 0 has no effect on the chromaticity, 100 has maximum effect.
6 Split Toning
This "partial color shift" (toning) is perhaps not the best name, but it must do...
We can not act directly on the color (hue) in the levels because the a* and b* channels are decomposed and it is near impossible to chart an accurate mathematical relationship between the chosen hue and such decomposed channels.
Nevertheless, we propose two curves built on the same principle as the chromaticity curve, where the x-axis represents the (discrete) number of levels (0-8), and the y-axis the force of the action with a positive effect above the equator and a negative one below it:
- The a* curve, which acts on the red-green tones,
- The b* curve, which act on the blue-yellow tones.
Note: if you apply identical adjustments to both a* ad b* curves, the end result will be identical to that which you would get had you used the chromaticity curve.
7 Edge Sharpness
This control is an addition to RawTherapee's existing arsenal of sharpening tools, including Unsharp Mask, RL Deconvolution, Edges, etc. Its principle is naturally based on the use of the wavelet decomposition. From a distance it vaguely resembles the Unsharp Mask to the extent that the wavelet decomposition creates a residual image which can be likened to a mask. But there the comparison ends. The wavelet decomposition of levels using the "Daubechies" method is a kind of direct recognition of contours as it scans each level of detail. The first level, level 0, uses every second pixel to act on details of a "radius" of about 2 pixels; the second details of a "radius" of 4 pixels, etc.
If the algorithm is interested only in the smallest details ((??? we can select the radius in the UI, so this "smallest details" doesnt really make sense)), and if a value is placed in reference((???)) (I called "Radius") and compare it to the current level, we are able to adjust the increase in contrast only for these levels, with a significantly higher increase values to those obtainable with the 10 sliders "Contrast".((this whole paragraph needs rewriting, I can't make interesting/useful sense of it))
There are 3 sliders:((this section is a direct copy, i did not rewrite it because its meaning is unclear to me))
- Radius may need to call it otherwise?((why?)) Its size will allow to increase the effect of perspective((perspective?)). The increase is not linear. Its action is not zero if the cursor is at zero. Discriminant action is maximum (for low wavelet levels), for low radius values, without being able to predict the level of action.((???))
- Value controls the strength((then why dont we call it strength?)). Setting it to 0 effectively turns sharpening off.
- Threshold (Threshold) allows different breakdown levels higher (distribution) for low levels (0, 1..2) when moving the cursor to the right.((???))
The edge sharpness tool is very sensitive to Maximum Levels setting because in principle it is interested only in the smallest levels, which are always decomposed.
8 Gamut - Controls
8.1 Reduce Artifacts in Blue Sky
Digital images often have some mottle noise in the blue sky. The wavelet process, because it increases local contrast, can generate or amplify small artifacts. This checkbox introduces a median filter to reduce these artifacts.
Do not forget to use the Noise Reduction tool which is called upstream in the toolchain pipeline.
8.2 Hue Targetting/Protection
This control is linked to the Contrast and Chromaticity wavelet controls. Toning is not affected. Moving the slider to the left will target the effects of these linked tools at the selected hue range. Moving it to the right will target the effects at all hues other than the chosen range.
The threshold selector widget below the slider lets you specify which hues to target (or protect). By default, a range of skin hues is selected. The selection is in the L*a*b* space: 100*PI, + 100*PI.
8.3 Avoid Color Shift
Applies an action on the chrominance and luminance of the recomposed image in order to take into account the working colorspace and deliver the data via a relative intent within the gamut of the working space.
9 Residual Image
As a reminder:
- The residual image uses the same data format as the original image (source) before decomposition.((what does this mean really, of what use to the user is this?))
- It corresponds to the difference between the original image and the sum of the decomposed levels (the action on each level has no effect on the residual image).
- The more levels are used, the more the residual image deviates from the original image.
Modifying the residual image is one of the key points of the wavelet processing. It allows one to:
- Work on the shadows and highlights independently of details,
- Reduce contrast (and chroma) overall, to better perceive micro-contrast,
- Change chrominance to reduce artifacts which result from excessive action of the levels (e.g. to the skies),
Moving the Shadows and Highlights sliders to the right increases luminance in these areas; to the left - decreases. This action is comparable to that of the Shadows/Highlights tool in the Exposure tab. Beware though that it does not invoke the principle of highlight reconstruction.
The action of these sliders is influenced by the two threshold sliders:
- Shadows Threshold, it is set to 30 by default (on the 0-100 L* scale), which means that only luminance values around 30 and below will be influenced by the Shadows slider.
- Highlights Threshold, it is set to 70 by default (on the 0-100 L* scale), which means that only luminance values around 70 and above will be influenced by the Highlights slider.
Read the L* value from the Navigator panel to help you decide the correct threshold values.
You could use the Shadows/Highlights adjusters to:
- add bloom to bright objects,
- prevent highlights from clipping,
- lift shadows,
This is one of the key points of the treatment made possible by wavelets. It allows you to change the contrast of the featureless residual image (remember, the residual image has no details, as those are handled by the contrast sliders) while letting you adjust the contrast of the detail levels using the contrast sliders above. For example by moderately reducing the contrast of the residual image, this leaves more room to accentuate that of each level, and thus increase the perception of depth and relief.
The same principle as for contrast slider above applies to the use of the chromaticity slider, but in addition you can act selectively on colors by use of the Hue slider in conjunction with the Hue Targetting/Protection slider. By default a typical blue sky-colored hue range is selected, but of course you can change it to anything you like, for example to match the colors of a flower, or skin. The Hue Targetting/Protection slider lets you choose how you want the chromaticity change to affect the selected hue range:
- By moving it to the left, only the chosen hues will be affected by your chromaticity change, e.g. select green grass hues, increase the Chromaticity slider, and move this Hue Targetting/Protection slider to the left to increase the chromaticity of only grass-colored in the residual image.
- By moving it to the right, your chromaticity change will apply too all hues except those chosen by the Hue selector. So in the example in the previous point, if you were to move the slider to the right, the chromaticity of all hues except the grass-colored ones would increase.
This control is very useful in the prevention of over-saturation of human skin, a carrot look which our eyes are used to noticing.