Frequency decomposition in Photoshop step by step. Simple Frequency Decomposition Material That Will Make You a Specialist

UPD (April 2014):
If this article seems to you unsuccessful and / or not very detailed, or you simply prefer to “look” rather than “read”, I suggest that you read the post. It contains a synopsis of the online master class of the same name and a video recording of MK (total duration 5 hours 13 minutes), cut into parts corresponding to the sections of the synopsis.

Everything was invented before us.

This is a short introduction with a brief background to the issue. Just for general development. If you are only interested in the direct description of the technique, you can skip directly to the next section.

It all began in 1822, when the French mathematician and physicist Jean Baptiste Joseph Fourier published his work "Analytical Theory of Heat", in which he used the representation of functions by trigonometric series for calculations. The Fourier expansion has proven to be a very convenient and flexible tool. In the 19th century, these ideas were actively developed by mathematicians, and in the 20th they were adopted by radio technicians. Direct (allowing the signal to be decomposed into a frequency spectrum) and reverse (allowing the signal to be restored from its particular spectrum) Fourier transforms have become the cornerstone of signal processing.

Digital technology has given rise to the need for digital signal processing, and computers have provided tremendous opportunities for this. And here the Fourier transform is actively used. Many information compression algorithms (including JPG and MPEG), noise reduction systems and much more are built on it. Naturally, such processing has not bypassed digital images either. This method has long been known among professional retouchers and in the West its name revolves around the word "wavelet". Although in reality it is based on a simpler Fourier transform, so I call it frequency decomposition.

This technique has not advanced much to the masses. I doubt that there is a "conspiracy of professionals" who hide from the public " secret tricks". Most likely, the specialists were simply too lazy to write articles, and the “broad masses” were quite happy with portraits (as consonant with polish, and the results are similar). Personally, I "opened America through the window" when a year and a half ago I was dealing with blending modes and found an article by Alexander Milovsky "We only dream about Moire." I figured out the frequency decomposition on my own and only later found out that many specialists have been using it for a long time and tightly. No wonder - everything was invented before us.

Fourier round trip or analogy with sound.

For people who are not familiar with signal processing, it will be easiest to understand the meaning of this method by analogy with sound processing, which occurs in almost all modern systems playback. I believe that it is wiser to understand the principle once than to rewrite incomprehensible settings from books of "miracle values" many times. But, if you are already familiar with signal processing, or you are simply not interested in understanding it, skip to the next section.

"Round trip Fourier" is a jargon student name for the standard signal processing method: direct Fourier transform decomposes the input signal (for example, at the input to an amplifier) ​​into a spectrum - the resulting spectrum changes in accordance with the characteristics of the device (for example, low frequencies are amplified) - the opposite Fourier transform allows you to get an output from the new spectrum (in our example, the bass will sound louder). The beauty of this method is that you can change different frequencies separately, independently of each other. The simplest change is strengthening or weakening. Those who find the old technique remember the "LF" and "HF" knobs on the amplifiers. These are the low and high frequency amplifiers. These adjustments did not affect the mids, but allowed the bass and treble to be boosted or attenuated separately. V modern technology the spectrum is divided into more frequency bands. Separate control of each of them is provided by an equalizer. The iTunes player, for example, has a ten-band equalizer.

But we don't need such a detailed separation. Let's go back to the three bands and see how the sounds in them contribute to the melody.

Low frequencies, they are also bass. They set the general rhythm, or, one might say, the general form of the melody. This is easy to verify by completely disconnecting all speakers and listening only to the subwoofer. You will immediately grasp the basic rhythm, although you may not even understand what kind of song it is. Thus:
Low frequencies carry information about the general form.

In the middle frequencies lies the human voice and the basic sounds of most musical instruments. This is the main, most informative and detailed part of the sound. Mobile phones reproduce only medium frequencies, which does not prevent us from listening to the interlocutor or enjoying (and more often quite the opposite) all kinds of ringtones. In other words:
The main details are in the middle frequencies.

The highest frequencies contain the harshest sounds (e.g. cymbals drum kit). Some of the high harmonics of other instruments are also located there. With the high frequencies turned off, the sound becomes smoother, loses sharpness. That is:

Now let's discard preliminary explanations, leave only brief conclusions and imagine that we are talking about an image. From the point of view of mathematics and signal processing, there is practically no difference between sound and image: sound is a one-dimensional (dependence of acoustic pressure or voltage at the amplifier terminals on time) a non-periodic signal, and a picture is a two-dimensional (dependence of brightness on horizontal and vertical coordinates) non-periodic signal ... For an RGB image, there will be three such signals at once: the brightness in the Red, Green and Blue channels. So in the image:

Low frequencies carry information about the general form.
The main details are in the midrange.
High frequencies add sharpness.

If we can spread different frequencies of the image to different layers, we get the following.
The first layer (LF) contains the largest details, the general distribution of brightness and color, that is, the shape of the object.
On the second layer (MF) - medium and fine detail, which is hidden behind the expression "local contrast" and for the close-up objects is a surface texture.
On the third layer (HF) - super-fine detail, which is often referred to as "micro-contrast" and which is responsible for sharpness.

It should be noted right away that boundary transitions are super-fine details, even if these are the boundaries of fairly large image elements. By the way, if you haven't guessed yet, the classic sharpening with the Unsharp Mask filter is just a strong boost in the high frequencies, and the sharpening with the HiRaLoAm method (High Radius Low Amount) is a weak boost in the mid and high frequencies.

How to do it in a graphics editor.

"Fully automatic" this can be done using plugins. For GIMP "and there is Wavelet decompose, for ImageMagick users there is a fairly detailed description (with code), those working in Photosop can use the 2D Fast Fourier Transform plugin for Adobe Photoshop or search the Internet for" wavelet "or" FFT ". I don't like plugins (there are reasons for this, but this is beyond the scope of this article), so we will do everything using standard Photoshop tools.

Let us take the simplest case, division into two components. The task is divided into two: first, to divide the picture into frequency components; second, add these frequency components back to the picture. To solve the first, we need low and high pass filters. The low pass filter leaves only frequencies below a certain one. Naturally, frequencies are not cut abruptly. Just starting from a certain value, they gradually weaken. A high-pass filter on the same principle leaves only frequencies above a certain one. These filters must complement each other. That is, after high-pass filtering, we should see the difference between the original picture and its low-frequency component. Otherwise, we will not be able to fold the image back.

The low pass filter in Photoshop is Gaussian Blur. The larger the radius we set, the lower the frequencies it leaves. An additional high-pass filter is High Pass (by the way, English term high-pass filter is translated as "high-pass filter"). Thus, after applying Gaussian Blur to the copy of the original image, we will get its low-frequency component. And after applying High Pass to the copy of the original image with the same Radius value, it is high-frequency, that is, the difference between the original image and its blurred copy. This difference is equal to the deviation of brightness from mid-gray (tone 128).

What are the reasons for choosing Radius. It is not very convenient to talk about the picture in terms of “frequency”, so let's move on to the terms “dimensions”. Decreasing the frequency corresponds to increasing the wavelength, and for the picture, this means increasing the size of the elements. The higher the Radius value, the larger size must have an element of the image to remain on the blurred copy, the more detail will disappear from it, leaving only the "clean" shape of the object. Thus, when using Gaussian Blur, we are interested in making the Radius as large as possible.

On the other hand, the larger the Radius value, the larger the size of the elements that the High Pass filter can have, the more information about the shape will pass into the layer on which we were going to leave only detail. Thus, when using High Pass, we are interested in keeping the Radius as low as possible. Since, according to the conditions of the problem, this value for Gaussian Blur and High Pass should be the same, a certain compromise option is selected based on the specific picture and the task at hand.

All work can be done in one file, and filters can be applied to copies of the layer with the original image. To make it easier to adjust the Radius value, I first convert these copies to Smart Objects (Convert to Smart Objects). Thus, filters are applied to them as smart filters and can be changed at any time. The illustration shows the appearance of the photo and the Layers palette for the original image, its low-frequency component (Low layer) and high-frequency component (High layer). The Radius value for both filters is 10.

Now the second half of the task: to add the received LF and HF components into a single picture. To do this, you need to change the brightness of each point of the blurred copy (low-frequency component) just as much as the brightness of the high-frequency component differs from the average (tone 128). We could have used the Linear Light blending mode, but its algorithm is slightly different: the brightness of each point of the lower layer changes by an amount twice as large as the difference between the brightness of the upper layer and the middle one (tone 128). Mathematically: R = S + 2 (C - 128), where S is the original brightness, C is the superimposed brightness, R is the resulting brightness. To compensate for the twofold gain set in Linear Light, it is necessary to halve the contrast of the high-frequency layer before adding it, leaving the average brightness unchanged. This can be done by creating a Curves Adjustment Layer on top of it, in which the black point is moved up to a value of 64, and the white point is shifted down to a value of 192.

Now the adjustment and high-frequency layers can be merged into a group and assigned a Linear Light blending mode to this group. Done, before you the original picture. If you want to minimize the total round-off error (this makes sense when decomposing into more components), work in 16-bit mode. But when decomposed into two components, it is so small that 8-bit sources may not be translated into 16 bits.

A few additional remarks. They do not affect the algorithm described above, so if you do not want to overload your brain, you can safely move on to the next section.

First remark. Alexander Milovsky suggested using Brightness / Contrast with Contrast = -50% to lower the contrast of the high-frequency layer. If you decide to do the same, remember to check the Use Legacy box. Since this article was written, the Brightness / Contrast algorithm has been changed and this checkbox makes it work according to the old algorithm. This can be clearly seen in the illustrations by K.

Second remark. On the forum site www.modelmayhem.com there is a post with the cute name "HighPass smokes for a long time (+ solution)", which claims that High Pass does not give an accurate result and the following method is proposed:

1.make two copies of the original image layer
2.apply to the bottom Gaussian Blur with the desired radius
3. Apply the Apply Image command to the top layer, choosing a blurred layer as the overlay image, the blending mode - Subtract, in the Offset field, enter 128, in the Scale field, enter 2.
4. enjoy the result, which "does not smoke for a long time".

Forgive me the author of this message, but I consider it nonsense. The described technique with the Scale 1 coefficient completely repeats the algorithm of the HighPass filter, and the Scale 2 coefficient simply halves the contrast (we did it with curves). Thus, shamanic dances with a tambourine lose all meaning. I tried to decompose and fold the picture using different methods and got exactly the same results. In addition, the immediately reduced contrast of the high-frequency layer will ruin our life during retouching. The same applies to the second algorithm proposed by the author. Those wishing to argue please in the comments.

Third remark. Sometimes instead of curves to compensate for the gain introduced by Linear Light, it is proposed to reduce the opacity of the high-frequency layer to 50%. In special cases, such a replacement may be equivalent, but in general it is not. In areas of light details on a light background (or dark details on a dark background), posterization may begin after an unweakened Linear Light blend, and reducing the opacity will not save you from it.

Why fence a vegetable garden?

The answer to this question is obvious: if we decomposed the image into its components, then now we can edit each of them separately.

For retouching a shape, it is most convenient to use a stamp with a reduced opacity. This allows you to smoothly smooth out the light and shadow pattern. But when retouching the face, the texture of the skin keeps us from doing this: it begins to blur and the face turns into a latex mask. Frequency decomposition solved this problem. Only the shape remained on the low-frequency layer, the entire texture of the skin went to the high-frequency layer. Therefore, you can safely work with a translucent stamp, brush and even your finger (Smudge Tool). You can smooth the shape. Thus, it is easy to remove uneven makeup, blemishes in the setting of light, reduce the depth of wrinkles, remove inflammation (in the illustration, I removed the redness from the scratch on the forehead, saving it). On selected sites you can not only smooth out, but also redraw the light-shadow pattern (this is how bags under the eyes are removed). With your finger, you can stretch, smoothly lengthen the existing shadows.

I did some retouching on a new transparent layer, turning on the Sample: Current & Below mode for the stamp. In the picture below, the areas edited on the low-frequency layer are marked in green (this is a specially made preview, with retouching such a picture will not be). The following is the appearance of the touch-up layer on a transparent and white background.

The Healing Brush does a good job of retouching textures, but when working on a whole image, it does not do very well to match the brightness of the original and the retouched areas. As a result, the image remains unpleasant spots... When working on the HF layer, this problem is removed. Note the large old scar on the forehead on the right. I took samples for retouching both to the left (where the forehead is decently lighter) and to the right (where the forehead is much darker) of him. But all the differences in brightness and color remained on the lower low-frequency layer, so the working conditions for the Healing Brush were just hothouse.

As in the previous case, retouching was done on a new empty layer with the Sample: Current & Below tool setting. Before retouching, switch the high-frequency group to the Normal (or Pass Trough) blending mode and turn off the contrast-reducing curves layer. A more contrasting picture makes it easier to navigate and work more accurately. In the picture below, the areas edited on the high-frequency layer are marked in green. The following is the appearance of the touch-up layer on a transparent and white background.

Since this is a technical illustration, no "glossy" licking has been done (I am generally not a supporter of polishing all faces under one cover). Several characteristic defects have been corrected as a sample. To see all the subtleties, you need to look at the full-size file. For those who cannot (or do not want) to download a large psd, below are two and a half times reduced before-after pictures. If the difference is not noticeable, you can pull them to your computer, lay them in two layers on top of each other and blink.

UPD: For those in the tank. I don't care how many pimples you think I have not cleaned up. This is a demonstration of the technique. Who wants to admire other people's results instead of learning, can look for other articles.

In the most critical cases, you can simply graft skin from one face to another. I once had to retouch a photo of a girl with a large strand of hair scattered all over her face. The task was simple: to remove all hair from the face, to make it as if it were alive. After wiping off the hair, the face turned into a latex mask (almost pure low-frequency component). A couple of her own photographs from a similar angle helped out. The HF component was taken from the "donors" and transplanted into the processed image with several flaps. In this operation, it is not necessary to use a photo of the same model (I took these photos simply because they were at hand). You can transplant skin from any face to any.

Decomposition into more components.

It makes no sense to compete with shumodavs (and some of them work exactly according to this principle) and decompose the picture into a very large number of components. But in some cases, it seems reasonable to decompose the picture into three components.

The decomposition into three components can be useful when dealing with defects of uniform size. For example, with freckles. On the low-pass layer for the Gaussian Blur filter, set minimum value the Radius parameter at which the freckles already disappear. On the high-frequency layer for the High Pass filter, set the maximum value of the Radius parameter at which freckles do not appear yet.

The mid-frequency layer is in the middle. This is a copy of the original image to which the High Pass filter is applied with a radius value like the Gaussian Blur of the LF layer, and then a Gaussian Blur with a radius value like the High Pass of the HF layer. Then a lowering contrast curves adjustment layer is applied to it and the entire group is assigned the Linear Light blending mode. The final layer structure is shown in the figure below.

Thus, freckles remain on the middle layer and are easily retouched. In areas where they are located very closely, it may also be necessary to slightly adjust the low-frequency layer.

Simplified method of application.

It is based on subtraction and is only suitable for smoothing the shape of the face ("killing the volume").

Apply Gaussian Blur to the copy of the original image with such a Radius value so that all the details interesting for you are gone from the picture. After that, apply a High Pass with such a Radius value so that only those elements that you think are excessive in volume (bags under the eyes, cheekbones, deep wrinkles, skin folds, etc.) are preserved in the picture. Invert the result (if you are working with a smart object, invert the image inside it) and change the blending mode to Linear Light (in this case, you don't even need to apply contrast-reducing curves). Mask the entire layer (Add Layer Mask \ Hide All) and use a white semi-transparent soft brush to paint on the mask those areas that you want to smooth.

Landmarks and promising directions.

I do not like to give specific settings for the tools, since each picture requires its own values. But as a general guideline, I can give you approximate Radius values ​​at which the corresponding elements go to the HF layer (for a breast portrait 12-25 megapixels).

Radius 1-2 - fine texture of the skin
Radius 4-5 - full texture of the skin, excluding deep wrinkles and similar defects
Radius 10-12 - full texture of the skin with large defects
Radius 25-30 - almost all local highlights and shadows (bags under the eyes, nasolabial folds, second and other chins, etc.)

In addition to the explicit and repeatedly described ideas (for example, the ability to change the global contrast while keeping the local one), I see several more ways for this processing method to develop. I myself will hardly be able to do this in the near future, if you want, you can experiment in this direction on your own. It will be interesting to know the results

1. Makeup. Frequency decomposition makes it easy to deal with the blemishes of the work of a makeup artist. But an extreme case of such a defect is a complete lack of makeup. You can try to superimpose it, working on the low-frequency component. It is possible that the HF will have to be modified somehow. It is better to develop specific methods with a photo with makeup as a sample (preferably in the same light, in the same angle and with the same model).

2. Fast and soft dulling of the texture of the skin while maintaining the rest of the details without waving a brush over the mask. Working on the high-frequency layer, use skin textures that contain little, but a lot of other details, red (RGB) and black (CMYK) channels for mixing and blending within the high-frequency group.

3. Shave the man. It doesn't take a lot of intelligence to transplant skin from a shaved chin. But to gently "shave off" one or two day stubble without resorting to the help of "donors" is an interesting task. Possible options solutions are a combination of the two previous examples.

In the end, I want to remind you. This technique is not a way of retouching skin and not even a technique for portrait retouching. It is versatile and suitable for any image where you need to sharply edit parts of various sizes.

A PR spoon in a barrel of knowledge.

Professional color correctors and people who easily fall into grumbling can safely skip this section.

Those wishing to attend face-to-face classes in color correction and image processing can get acquainted with the programs and a list of upcoming events in. There you will also find links to my other articles.

Without prior agreement with the author, it is allowed to reprint and post this material on any resources with free access, provided that the text (including this section), links and illustrations, attribution and links to the first publication are fully preserved.

For commercial use or reprints with amendments must be agreed with the author. You can contact me by e-mail site

When processing a portrait, we all face a problem - how to remove skin imperfections, make the skin smooth and velvety, as well as preserve its texture. I think many of you have already heard about such a Photoshop plug-in as Portraiture - this, as they say, is for especially lazy ones :) Fast, convenient, but it makes the skin unnatural and plastic. I personally sometimes use it, but extremely rarely and with high transparency, it is convenient for full-length photographs, where the face is quite small. But let's talk about a more professional skin retouching method, which, while tedious and slower, can achieve superior results.

So, we have a photo on the left, in which the primary light and color correction in lightroom has already been done. After our manipulations, we get the photo on the right.

There are quite a few different ways this method of retouching, but I will tell you which one is faster, more convenient, simpler and gives the best result.

What this method is based on: it is based on the fact that any picture can be decomposed into 2 components - high and low frequencies. And if we "fold" them back - we get the original picture. Thus, we divide the image into a layer that contains the texture of the skin and a layer that contains information about color, light-shadow transitions and "shape". The low pass filter in Photoshop is the Gaussian Blur filter. Upper - High Pass, which we will apply through the External channel, because the filter itself gives inaccuracies when "gluing". It sounds complicated, but I'll try to explain simple language and examples)

Let's move on to practice:

1. Let's say we have a background layer. Make 2 copies of it - I do it using the Ctrl + J hotkey. You can go to the Layer menu - Duplicate Layer ...

1. Rename them - the lower copy is called low, the upper one - high.

1. Turn off the visibility of the top layer high by clicking on the eye icon.

1. Select the low layer, now we only work with it. Apply a Gaussian Blur filter to it. To do this, go to Filter - Blur - Gaussian Blur ... This creates a layer that stores the “color and shape” of the image.

1. What radius should we choose? It is important to choose such a blur radius so that the skin texture is no longer visible, but all light-shadow irregularities from skin imperfections are preserved. Such a radius ranges from 3px (where there is a small face area / full-length photo - or a small photo size) to 15px (a large portrait or beauty, big size Photo). In my case, I was quite happy with the result at 3.5px, because I cropped the photo.

1. Select the high layer and turn on its visibility.

1. Now we are working with this layer, creating the "texture" of the image. To do this, go to Image - Apply Image ... (Image - External channel ...)

1. In the Layer tab, click and select our blurred layer - low.

1. IMPORTANT:

1 case - if you work in 8-bit mode, then set the settings - Blending - Substract, Scale - 2, Offset - 128.

2 case - if you work in 16-bit mode, then set the settings - Blending - Add, Scale - 2, Offset - 0. And click the Invert checkbox.

How to see how many bit channels you are working on? This is usually written at the top next to the name of the open image separated by a slash (for example, Beauty.psd (… / 8) or (… \ 16)). If you do not see, then go to Image - Mode - in the open tab, the mode in which you are currently working will be marked with a tick.

1. We see a gray image. Change the blending mode of the high layer to Linear Light (Linear light).

1. I suggest grouping these two layers. Select them and press Ctrl + G or go to Layer - Group Layers (Layers - Group Layers). This group is called “frequency decomposition”.

Try to turn off / on the visibility of this group - you will understand that the image does not change, we decomposed the image into 2 components - into the lower frequencies - a blurred layer, which contains information about the color and shape, and to the upper ones - where only the texture of the image is stored. Now we can work with these layers separately, without fear of "screwing up").

First, we fight all skin irregularities - acne, scars, moles, excess hair - information about them is stored in the texture layer, i.e. on the high layer. We will work with the Stamp tool, with a hardness of 100%. Why? Because this is the texture of the skin and “blurred edges” that remain after working with a soft brush or a healing brush, or the same patch are unacceptable. We select the size a little larger than the pimple.

Made? Now select a good texture next to the pimple, i.e. smooth. Press Alt and click. Thus, we have chosen the place from where we will take the texture. Now we boldly click on the pimple itself. Fine!

So we go all over the skin, removing everything that we don't like)

We replaced the bad texture with a good one, but the spots from these pimples and irregularities still remained. The lower layer is already responsible for them, we will work with it.

To do this, select the low layer. We need the Mixer Brush Tool with 10% at the top.

What does he do? The name itself is the essence of his work - he mixes nearby colors wherever you brush. Thus, moving a little with a brush at the place of the pimple and in its vicinity, we will mix the color so that in its place there will be Smooth surface instead of a stain. Try it.

1. Thus, we go over all the spots on the face, leveling the texture of the skin and creating an even complexion, but do not overdo it.

This is basically the whole method. frequency decomposition)

To achieve better result, there are several improvements to this method, which I will now tell you about.

1. Softening the skin.

To do this, before you work on the low layer, make a copy of it and apply a Gaussian Blur filter to the copy with a blur of no more than 10px, so that you can clearly see how the skin softens.

Now let's create a layer mask. To do this, click on the corresponding icon.

Now we take a white brush and paint where we need softening, namely the forehead, cheeks, nose, chin, but do not go over the eyes, lips or the edge of the face. As a result, the mask will be something like this:

I lowered the opacity to 65%, you can do it on your own)

1. Sharpness and better vision of all flaws in retouching.

We do this before we started working with the texture.

Make a copy of the high layer by pressing Ctrl + J. Create a clipping mask for it. To do this, select the high copy layer, hold down Alt and move the cursor to the position between these layers, until such an icon appears. We click.

Change its Blending Mode to Normal.

Now we will carry out all actions with the stamp on it.

To better see all the imperfections, let's create an auxiliary curves layer. Click on the icon and select Curves.

Rename this layer to contrast, as this layer really enhances the contrast. And create a clipping mask for it, as we did above.

For the bottom point: Input - 97, Output - 0.

For the top point: Input - 158, Output - 255.

We see how this affects our image:

It has become too sharp, so lower the Opacity to about 50% so that we can clearly see all the irregularities.

When you finish retouching, reduce the opacity to about 1-5%, if you want to sharpen the entire image, if you like the original image, then either turn off the visibility of this layer, or delete it altogether. That's all)

P.S. I am still not happy with the result as some spots are visible on the skin. To remove them, I use the lightening and darkening methods, popularly known as the Dodge and Burn method, which I talked about. After that, we get a much more attractive picture. Then I did the final color correction and here is the result: the original picture, after frequency decomposition and the final one.

When processing a portrait, the question often arises of how to preserve the texture of the skin, getting rid of all the irregularities and imperfections. There is a Portraiture Photoshop filter that makes the skin look plastic. It is not right. No matter how much it is said that it should be used in moderation and dosed, the result of its use leaves much to be desired. We will now talk about how to achieve a good result. This is not the way for the lazy.

Frequency decomposition is a retouching technique that has been around for a long time. It is difficult to learn, but in the future it is quite easy to apply it in practice. All decomposition actions can be recorded into an action, and then the processing process is automated and simplified as much as possible. This article will walk you through the entire layer decomposition process step by step.

Let's get started. Create a copy of the background and go to the Filter-Other-High Pass menu. The filter turns the layer gray. Set the settings to 6 px. The value can be changed in the range 2 - 11 px. The choice of settings depends on. how large the texture of the skin is. When shooting a shabby close-up, the value should be set from 9 to 11 px. If the model is photographed waist-deep or full-length, then the texture of the skin will be small in the picture. In this case, the value should be set to 2-4 px. In this method, it is very important to choose the right filter value. Too low a value will not render the texture of the skin properly, and too high will make it rough.

Having decided on the settings, click OK and rename the layer to High Pass. Now, for convenience, create a group (Ctrl + G), place the processed layer there and name the group - "Texture"
Now change the blending mode of the whole group to Linear light.

The picture immediately became colored, but the detail is too high. Let's fix it now.

Select the High Pass layer in the group with the mouse and create a Curves adjustment layer above it.

In the settings, we reduce the slope of the curve as in the screenshot. Now you can close this window.

Now make a copy of the background layer again. Apply Filter-Blur-Gaussian Blur to it and set the value to 6 px in the settings. It is very important that the values ​​in the previously applied High Pass filter and in Gaussian Blur must be the same radius setting. Apply the filter and change the name of the layer to "Blur".

This layer is also placed in new group, which we will call "Spot" (spots).

Now you can see that the picture has taken on its original form. If you turn off all the created layers, then you can see that there is no difference. (Layers can be quickly turned off by holding Alt + click on the eye. This action will turn off all the layers that are above.) The meaning of all these manipulations is that in the upper group we got the texture of the image, and in the lower group you can draw, blur and adjust basis. All actions in the lower group will not affect the texture.

Now, in order not to be distracted by the texture, turn it off. Let's go to the bottom group. We see spots and irregularities. Create a copy of the Blur layer and apply Gaussian Blur. This time we define the blur by eye. Smear until small and medium spots disappear. Do not touch large spots the size of a nose or a voice. Blurring is most often done with a value between 10 and 25 px.

Create a mask for this layer and fill it with black.

Draw a face with a white brush, without touching the eyes, lips, nostrils and other contrasting areas. The more blur is applied, the further you need to stay away from borders, contrasting colors. Otherwise, they will float.

Select the brush and set the Opasity and Flow parameters to 10%. It is desirable that these parameters are equal, since the brush works more correctly in this case. Using an eyedropper, take a color sample from the area that you like the most. With a brush, paint over the spot in the area that you don't like. This technique can be used to remove glare from flash or circles under the eyes. When drawing, it is very important to keep an eye on it. so that the areas that are being sketched do not become flat. For high-quality retouching, you should have at least some art education. It also does not hurt to be like the courses of makeup artists.

After drawing all the problem areas in the "Spot" group, we move on to work in the "Texture" group. Turning on the group with the texture, we can see that the texture is preserved, and the skin is evened out.

With all this, it is clearly visible that the texture in the places of the former flare from the flash and where there were circles under the eyes, has become too rough.

Create a copy of the High Pass layer and take the Healing Brush Tool. In the brush settings it is better to make it oval. This will make the effect of the brush less noticeable. Set the hardness of the brush to about 90%. In the brush settings, you MUST set Current Layer. Then we work with a brush as usual. With Alt we take a normal area and work on the "sick" one. If, after working with a brush, the borders remain too sharp, then you need to apply a white mask, and wipe the necessary places with a black brush with a slight transparency.

If there are places with a weak texture, then you need to go to the adjustment layer Curves 1 and on the layer mask with a black brush with Opasity and Flow equal to 20%, paint the places with a weak texture. The texture of the skin in these areas will be enhanced.

At the end of the processing, you need to reduce the transparency of the "Blur copy" layer to 70 - 80%. A value of 100% makes the skin look too perfect. This does not happen.

The article describes only one stage of portrait processing. The pictures below show four processing options: 1 - converting from RAW without changes; 2 - a consequence of the primary processing in the RAW-converter, and the drawing of light and shadow; 3 - the result of frequency decomposition; 4 - the final result with the elaboration of color and contrast.

Let's look at a few more portraits that have been processed by the frequency decomposition method.

Based on materials from the site:

16.04.2017 3362 Feature articles 0

An important part in retouching portraits is the detailing of the skin. One of the most convenient processing methods is the frequency decomposition method. It is very simple, if you work with it clearly according to the instructions, and subsequently bring this algorithm of action to automatism.

The frequency decomposition method works flawlessly, even in the most difficult situations when the photographer does not know where to start. For retouching portraits, it is important to protect small parts... Therefore, it is better to use the frequency decomposition method than processing with smart tools.

PREPARATION

To begin with, we will carry out a number of preliminary mandatory procedures with the image. Let's divide it into the same frequencies, component parts. Let's get started.

1. First of all, create 2 layers, copies of the original image. This can be done in different ways: for example, by right-clicking and selecting "create a duplicate layer" or by pressing the keyboard shortcut Ctrl + J. Let's immediately name the layers and call them low and hi, later these will be our low and high frequencies.

Making copies of a layer is a very simple and useful habit, especially when retouching. You will always have the opportunity to refer to the source if something goes wrong.

2. Turn off the visibility of the top layer hi, we will deal with it later. Move on to the low layer.

3. This hi layer must be blurred using the Gaussian Blur filter, in which you must select the blur radius. We need to remove skin irregularities, fine wrinkles, dust and the like. But the eyes, nose and eyebrows, we most likely want to keep. Therefore, starting to increase the radius, we carefully look at how the picture changes. The result is highly dependent on the size of the source, as well as on how much space the person takes in the frame. For this image, let's stop at 12.5px for the blur radius. It is important to remember the selected value, this number will be useful to us in the next step.

Over time, you will learn to determine the appropriate radius right away.

4. Now go to the top hi layer. We need to compensate for the image so that when overlaid on our blurred low layer, we get the original image as a result. To do this, a filter is used, which in combination with a suitable blend mode will give an effect that is the opposite of the Gaussian Blur that we used earlier. And the name of this filter is High Pass (Filter - Other - Color Contrast).

In the High Pass window that opens, there is only one adjustable parameter, and this is again the radius. Here you need to set the value from the last step. Set it to 12.5px, which is how much we blurred the low layer.

5. Set the hi layer to Linear Light blending mode.

This mode works in such a way that the gray color becomes transparent, and any deviations from the gray dramatically increase the brightness and contrast. The result after overlaying on a blurred layer is exactly twice as high-contrast than the original image.

6. You can reduce the contrast in different ways, I prefer using adjustment layers, that is, layers with settings (for example, a layer with curves). This method easily allows you to temporarily disable the contrast reduction. It is very convenient to work with the contrast version during retouching, all irregularities and details that can be accidentally missed when looking at the original version are immediately highlighted.

So, create an adjustment layer: New Adjustment Layer - Curves (Layers - New Adjustment Layer - Curves).

7. In order to apply a reduction in contrast only to our top layer, create a Clipping Mask (Layers - Layers - Create Clipping Mask).

You can also do this by holding Alt and clicking on the joint between the layers for which we want to create a Clipping Mask. At this moment, the cursor will change the icon to an arrow with a square, and after clicking in front of the layer, only an arrow will remain. This means that everything went well.

Go to the adjustment layer settings (they appear by double-clicking on the curve image in the layers palette). Here we are interested in the extreme points, they are responsible for the boundaries of the image brightness: the lower left is the blackest, the upper right is the whitest. To reduce the contrast, drag the bottom up and the top down. To accurately hit the value and reduce the contrast by half, you need to tighten these points by exactly a quarter from the top and bottom, respectively.

Everything came together, the picture on the screen looks exactly the same as the original original. Our image is split in two and ready to go.

The high frequency layer, which we call hi, contains the texture of the skin, fine details of clothing and hair. But hair color and skin color are in the low layer. Thus, it is convenient to correct skin irregularities and defects in the layer with a high hi frequency.

HIGH FREQUENCIES

Turn off the visibility of the curves adjustment layer and change the blending mode of the gray hi layer back to Normal. Create a new layer on top of the hi, and it automatically switches to the Clipping Mask. This layer will be our retouching.

Be sure to 100% opacity and 100% hardness of the brush (adjusted together with the size of the brush by right-clicking), so that blurry areas do not appear. We ignore the color on this stage, so all the textures fit perfectly with the hard stamp settings.

Finally, we move on to the actual skin retouching. And this is the usual work with a stamp: we take an area with a texture you like and replace what is required. Moles are not defects, but everything else needs your attention.

RESULTS

In my opinion, the picture as a result turned out to be quite natural. Skin texture, color shades, brightness differences and some defects in the portrait have undergone a detailed study.

You already saw the result at the beginning of the lesson:

That's all. All you need is photographic, friends!

06.12.15

When processing a portrait, we all face a problem - how to remove skin imperfections, make the skin smooth and velvety, as well as preserve its texture.
I think many of you have already heard about such a Photoshop plug-in as Portraiture - this, as they say, is for the especially lazy 🙂 Fast, convenient, but it makes the skin unnatural and plastic.

But let's talk about a more professional skin retouching method, which, while tedious and slower, can achieve superior results.

So, we have a photo on the left, in which the primary light and color correction in lightroom has already been done. After our manipulations, we get the photo on the right.

There are quite a few different ways of this retouching method on the net, but I will talk about which one is faster, more convenient, simpler and gives the best result.

What this method is based on: it is based on the fact that any picture can be decomposed into 2 components - into high and low frequencies. And if we "fold" them back - we get the original picture. Thus, we divide the image into a layer that contains the texture of the skin and a layer that contains information about color, light-shadow transitions and "shape". The low pass filter in Photoshop is the Gaussian Blur filter. Upper - High Pass, which we will apply through the External channel, because the filter itself gives inaccuracies when "gluing". It sounds complicated, but I will try to explain it in simple language and with examples)

Let's move on to practice:

1. Let's say we have a background layer. Make 2 copies of it - I do it using the Ctrl + J hotkey. You can go to the menu Layer - Duplicate Layer ...

2. Rename them - the lower copy is called low, the upper one - high.

3. Turn off the visibility of the top layer high by clicking on the eye icon.

4. Select the low layer, now we only work with it. Apply a Gaussian Blur filter to it. To do this, go to Filter - Blur - Gaussian Blur ...

This creates a layer that stores the “color and shape” of the image.

5. What radius should we choose? It is important to choose such a blur radius so that the skin texture is no longer visible, but all light-shadow irregularities from skin imperfections are preserved. This radius ranges from 3px (where there is a small face area / full-length photo - or a small photo size) to 15px (large portrait or beauty, large photo size). In my case, I was quite happy with the result at 3.5px, because I cropped the photo.

6. Select the high layer and turn on its visibility.

7. Now work with this layer, creating a "texture" of the image. To do this, go to Image - Apply Image ... (Image - External channel ...)

8. In the Layer tab, click and select our blurred layer - low.

1 case - if you are working in 8-bit mode, then set the settings - Blending - Substract (Overlay - Subtraction), Scale (Scale) - 2, Offset (Shift) - 128.

2 case - if you are working in 16-bit mode, then set the settings - Blending - Add (Overlay - Adding), Scale (Scale) - 2, Offset (Shift) - 0. And click the Invert checkbox (invert).

How can you see how many bit channels you are working on? This is usually written at the top next to the name of the open image, separated by a slash (for example, Beauty.psd (… / 8) or (… \ 16)). If you do not see, then go to Image - Mode - in the open tab, the mode in which you are currently working will be marked with a tick.

10. We see a gray image. Change the blending mode of the high layer to Linear Light (Linear light).

11. I suggest grouping these two layers. Select them and press Ctrl + G or go to Layer - Group Layers (Layers - Group layers). This group is called “frequency decomposition”.

12. Try to turn off / on the visibility of this group - you will understand that the image does not change, we decomposed the image into 2 components - into the lower frequencies - a blurred layer, which contains information about the color and shape, and to the upper ones - where only the texture of the image is stored. Now we can work with these layers separately, without fear of "screwing up").

13. First, we fight all skin irregularities - acne, scars, moles, excess hair - information about them is stored in the texture layer, ie. on the high layer. We will work with the Stamp tool, with a hardness of 100%. Why? Because this is the texture of the skin and “blurred edges” that remain after working with a soft brush or a healing brush, or the same patch are unacceptable. We select the size a little larger than the pimple.

14. Did you do it? Now select a good texture next to the pimple, i.e. smooth. Press Alt and click. Thus, we have chosen the place from where we will take the texture. Now we boldly click on the pimple itself. Fine!

15. So we go all over the skin, removing everything that we do not like)

16. We replaced the bad texture with a good one, but the spots from these pimples and irregularities still remained. The lower layer is already responsible for them, we will work with it.

To do this, select the low layer. We need the Mixer Brush Tool with 10% at the top.

What does he do? The name itself is the essence of his work - he mixes nearby colors wherever you brush. Thus, by brushing a little on the spot and in its vicinity, we will mix the color so that in its place there will be a flat surface instead of a spot. Try it.

17. Thus, we go over all the spots on the face, leveling the texture of the skin and creating an even complexion, but do not overdo it.

This is, in principle, the whole method of frequency decomposition)
To achieve the best result, there are several improvements to this method, which I will now tell you about.

1. Softening the skin.

To do this, before you work on the low layer, make a copy of it and apply a Gaussian Blur filter to the copy with a blur of no more than 10px, so that you can clearly see how the skin softens.

Now let's create a layer mask. To do this, click on the corresponding icon.

Now we take a white brush and paint where we need softening, namely the forehead, cheeks, nose, chin, but do not go over the eyes, lips or the edge of the face. As a result, the mask will be something like this:

I lowered the opacity to 65%, you can do it on your own)

2. Sharpness and better vision of all flaws in retouching.

We do this before we started working with the texture.

Make a copy of the high layer by pressing Ctrl + J. Create a clipping mask for it. To do this, select the high copy layer, hold down Alt and move the cursor to the position between these layers, until such an icon appears. We click.