Noises can be random analogue, pulsed and various types of deterministic.
Random analog noise
Random analog noise is generated, as a rule, by the granular structure of the photographic material in which the original was made. Noises become relevant at magnifications greater than 8x.
To eliminate such noise, smoothing filtering methods are used.
The operation of these methods is based on digital filtering by averaging the signal value over the vicinity of the read pixel. In programs like PhotoShop, these anti-aliasing filters are called Blur, Gaussian Blur.
Blur will give direct averaging. Gaussian Blur introduces pixel weights into the averaging matrix according to Gaussian law.
Blur is obsolete because it does not allow you to adjust the degree of averaging. The degree of smoothing is adjusted by repeated application of the filter.
Gaussian Blur is more modern. In it you can adjust the averaging parameter, thus adjusting the smoothing.
It must be remembered that the use of such filters can lead to loss of image sharpness, since not only the noise structure is averaged, but also the pixels that form the image boundary. In some cases, it is advisable to additionally carry out an unsharp masking procedure after the smoothing procedure.
Random impulse noise
Random impulse noise refers to relatively sparsely located single defects, such as scratches and dust particles. When applied to them, the smoothing procedure is usually not effective due to the fact that the size of such defects is quite large.
To eliminate such defects, rank-ordinal class filters are used. Such rank-order filters create a series of pixels along a line, order these series by placing them in ascending order, reject the minimum and maximum pixel values that may be defective, and find the average value in this series. This average value is put in place of the analyzed pixels.
In this way, relatively minor defects such as scratches and dust can be eliminated. In principle, it is possible to change the length of the series and thus select for larger defects.
However, for fairly large defects that exceed the length of the pixel series, this method is not applicable.
This is the method used by the Dust and Scratches filter.
For larger impulse noise, it is necessary to resort to semi-automatic retouching, in which image defects are eliminated by replacing defective pixels with colored pixels from their immediate surroundings. A pixel is selected from the immediate surroundings and placed on the defective location.
In software, this procedure is called a stamp and requires a significant investment of time. Before proceeding with such a procedure, it is necessary to analyze the image at a magnification scale during reproduction and eliminate those defects that will be noticeable at this scale. In principle, the same procedure can be used for editorial correction, when it is necessary to supplement some lost details of the image.
Deterministic Image Noise
The most striking representative of deterministic image noise is the raster structure of the image, if a printing print is used as the original.
Reading a raster image may result in unwanted interaction of the image's raster structure with the new raster structure generated during the photo output process.
There are two possible ways to solve this problem:
1. elimination of the raster structure of the original during scanning and processing. For this purpose, methods are used similar to the methods of aperture filtering when reading an image with a larger aperture, or their digital analogue, that is, pixel averaging and the formation of an average signal.
It has been shown theoretically and experimentally that best results are obtained by matching the size of the aperture with the dimensions of the raster element of the raster structure of the original. Therefore, during the scanning process, it is necessary to accurately determine the raster lineature that was used in the original and select the derasterization filter in accordance with this lineature.
To determine the raster lineature in the original, it is possible to use special tests. Some modern programs, for example LinoColor, allow you to determine the lineature during the preliminary scanning and, in accordance with it, set the optimal derastration filter.
Disadvantages of this elimination:
1) loss of image sharpness;
2) as a consequence different angles rotate raster image structures to different colors, complete coordination of the derastration aperture and the raster structure does not occur and residual fluctuations in the image (moiré formation) are inevitable.
2. reading the raster structure with its complete preservation. As a result, when reading three raster color-separated images, we will obtain the raster structure being preserved. In fact, we will receive an image in the SMUK system. This image can then be transferred to Lab, thus losing information about the raster structure. Then transfer all processing to Lab and go to SMUK with its raster structure.
To do this, it is necessary to read at high resolution. СopiDot – corresponding software to transfer SMUK to Lab.
The current difficulty is that reading color images has significant difficulties. Therefore, this kind of CopiDot technology is currently used to read rasterized and color-separated photoforms.
Particularly interesting is this technology, which Lately has developed greatly, with the need to use some kind of archival photo frames C-t-P technologies(computer-printed form).
3. Rerastering using a raster of irregular structure (frequency modulated).
The image can be damaged by noise and interference of various origins, such as video sensor noise, grain noise in photo materials, and errors in the transmitter channel. Their influence can be minimized using classical statistical filtering methods. Another possible approach is based on the use of other heuristic methods for spatial processing.
Video sensor noise or errors in the transmission channel usually appear in the image as discrete changes in isolated elements that do not have spatial correlation. Distorted elements often differ quite noticeably from neighboring elements. This observation has served as the basis for many algorithms that provide noise reduction.
The use of digital image filtering can significantly improve the quality of the image obtained during UWB sensing.
Next, we will consider the use of linear filtering to smooth out noise in the image (low-pass filtering), emphasizing the boundaries of objects using high-pass filtering, as well as the median filtering method for eliminating pulse-type noise.
Rice. 7 explains a simple threshold noise reduction method in which the brightness of all image elements is measured sequentially.
Rice. 3.7. Threshold noise reduction method. If brightness exceeds the average brightness of the group of closest elements by a certain threshold value, the element's brightness is replaced by the average brightness:
If 
]
Because noise is spatially decorrelated, its spectrum tends to contain higher spatial frequencies than the spectrum of a normal image. Therefore, simple low-pass spatial filtering can serve effective means noise smoothing. Array Q of size MM of the output image is formed by discrete convolution of array F of size NN of the source image with a smoothing array H of size LL according to the formula
Noise smoothing is provided by low-pass filtering using an H array with positive elements. Below are three types of smoothing arrays, often called denoising masks:
These arrays are normalized to obtain unity gain so that the noise reduction procedure does not cause a shift in the average brightness of the processed image. If the required noise reduction involves the use of large arrays, it is advisable to perform the convolution indirectly using the Fourier transform, since this usually results in a gain in the amount of computation.
Emphasizing boundaries.
In electronic image scanning systems, the resulting video signal can be passed through an electrical high-pass filter. Another way to process scanned images is to use unsharp masking. In this case, the image is, as it were, scanned by two overlapping apertures, one of which corresponds to normal resolution, and the other to reduced resolution. As a result, an array of normal image F (j, k) and an array of fuzzy image F L (j, k) are obtained, respectively. Then a masked image array is formed
F M (j, k) = c F (j, k) - (1-c) F L (j, k),
where C is the proportionality coefficient. Typically the C value ranges from 3/5 to 5/6, i.e. The ratio of the components is normal and reduced clarity varies from 1.5 to 5.
Edge enhancement can also be achieved by performing discrete filtering according to equation (1) using the high-frequency impulse response H. Below are three typical masks for performing high-pass filtering:
These masks differ in that the sum of their elements is equal to one.
Another way of emphasizing boundaries is the so-called statistical differentiation. The brightness value of each element is divided by the statistical estimate of the standard deviation (j,k)
G (j,k) = F (j,k) / (j,k).
Standard deviation
is calculated in some neighborhood N(j,k) of the element with coordinates (j,k). Function 
- the average brightness value of the original image at a point with coordinates (j,k), approximately determined by smoothing the image using the low-pass filtering operator according to formula (3.1). The improved image, represented by the array G (j,k), differs from the original image in that its brightness is higher at the boundaries, the elements of which are dissimilar to neighboring elements, and lower in all other areas. It should be noted that emphasizing useful boundaries is accompanied by an increase in noise components.
Median filter.
Median filtering is a nonlinear signal processing technique developed by Tukey. This method is useful in reducing noise in an image. The one-dimensional median filter is a sliding window covering an odd number of image elements. The center element is replaced by the median of all image elements in the window. Median of a discrete sequence
a 1 , a 2 , ..., a N for odd N is the element for which there are (N-1)/2 elements less than or equal to it in value, (N- 1)/2 greater than or equal to it in value size. Let the window contain image elements with levels 80, 90, 200, 110, 120; in this case, the center element should be replaced with the value 110, which is the median of the ordered sequence 80, 90, 110, 120, 200. If in this example the value 200 is a noise outlier in a monotonically increasing sequence, then median filtering will provide a significant improvement. On the contrary, if the value 200 corresponds to the useful signal pulse (when using wideband sensors), then the processing will lead to a loss of clarity in the reproduced image. Thus, the median filter in some cases provides noise suppression, in others it causes unwanted signal suppression.
The median filter does not affect step or sawtooth functions, which is generally a desirable property. However, this filter suppresses pulsed signals whose duration is less than half the window width. The filter also causes the vertex of the triangular function to flatten.
The ability to analyze the effect of the median filter is limited. It can be shown that the median of the product of the constant K and the sequence f(j) is equal to
med( K f(j) )=K med(f (j)).
Besides,
med( K+ f(j) )=K + med(f (j)).
However, the median of the sum of two arbitrary sequences f (j) and g(j) is not equal to the sum of their medians:
med( g(j)+ f(j) )=med(g(j))+ med(f (j)).
Various strategies are possible for applying the median filter to suppress noise. One of them recommends starting with a median filter, the window of which covers three elements of the image. If the signal attenuation is insignificant, the filter window is expanded to five elements. This is done until median filtering begins to do more harm than good. Another possibility is to perform cascaded median filtering of the signal using a fixed or variable window width. In general, those areas that remain unchanged after a single filter treatment do not change after repeated processing. Regions where the duration of the pulse signals is less than half the window width will be subject to changes after each processing cycle.
The concept of a median filter can easily be generalized to two dimensions by using a two-dimensional window of the desired shape, such as rectangular or close to circular. It is obvious that a two-dimensional median filter with a window of size LL provides more effective noise suppression than sequentially applied horizontal and vertical one-dimensional median filters with a window of size L1; 2D processing, however, results in more significant signal attenuation.
The median filter suppresses scattered impulse noise more effectively than smooth noise. Median filtering of images for the purpose of noise suppression should be considered a heuristic method. It cannot be used blindly. Instead, you should check your results to ensure that median filtering is appropriate.
Reduce noise in images
Quite often, when generating visual data, the resulting images are noisy. This is due to the imperfection of the equipment, the influence external factors and so on. The end result is a deterioration in the quality of visual perception and a decrease in the reliability of decisions that will be made based on the analysis of such images. Therefore, the task of eliminating or reducing the level of noise in images is urgent. A lot of work has been devoted to solving the problem of noise filtering; there are various methods and algorithms. In this work we will consider only some approaches and the possibilities of their implementation in the Matlab system.
Step 1: Reading the original image.
Step 2: Formation of noisy images.
Step 3: Using a median filter to remove impulse noise.
Step 4: Suppress the noise component using the smoothing operation.
Step 5: Threshold noise reduction method.
Step 6: Low-pass filtering using noise-reducing masks.
Step 1: Reading the original image.
We read the image from the file into working space Matlab and display it on the monitor screen.
L=imread("kinder.bmp");
figure, imshow(L);
Fig.1 Original image.
Step 2: Formation of noisy images.
In the Matlab system (Image Processing Toolbox), it is possible to generate and overlay three types of noise on an image. To do this, the built-in imnoise function is used, which is intended primarily for creating test images used in selecting and exploring noise filtering methods. Let's look at a few examples of applying noise to images.
1) Adding impulse noise to the image (by default, the noise density is equal to the fraction of distorted pixels):
L2=imnoise(L,"salt&pepper", 0.05);
figure, imshow(L2);
Fig.2. Noisy image (impulse noise).
2) Adding Gaussian white noise to the image (by default, the expected value is 0 and the variance is 0.01):
L1=imnoise(L,"gaussian");
figure, imshow(L1);
Fig.3. Noisy image (Gaussian noise).
3) Adding multiplicative noise to the image (by default, the expected value is 0 and the variance is 0.04):
L3=imnoise(L,"speckle",0.04);
figure, imshow(L3);
Fig.4. Noisy image (multiplicative noise).
Step 3: Using a median filter to remove impulse noise.
One of the effective ways to eliminate impulse noise in an image is to use a median filter. Most effective option is implemented in the form of a sliding aperture.
For i=1+n1:N+n1;
disp(i) for j=1+m1:M+m1;
Rice. 5. Restoring an image distorted by impulse noise using the median filtering method.
The restored image is only slightly different from the original image and is significantly better, in terms of visual perception, than the noisy image.
Step 4: Suppress the noise component using the smoothing operation.
There is a class of images for which suppression of the noise component can be achieved using a smoothing operation (low-frequency spatial filtering method). This approach can be applied to image processing containing regions large area with the same brightness level. Note that the level of the noise component should be relatively small.
F=ones(n,m); % n and m dimension of sliding aperture
Lser=filter2(F,Lroshyrena,"same")/(n*m);
Rice. 6. Restoring an image distorted by impulse noise using a smoothing operation.
The disadvantage of this method, unlike the median filtering method, is that it leads to blurring of the boundaries of image objects.
Step 5: Threshold noise reduction method.
Image elements that have been distorted by noise appear noticeably different from neighboring elements. This property formed the basis of many noise suppression methods, the simplest of which is the so-called threshold method. When using this method, the brightness of all image elements is sequentially checked. If the brightness of a given element exceeds the average brightness of the local neighborhood, then the brightness of this element is replaced by the average brightness of the neighborhood.
For i=1+n1:N+n1;
disp(i) for j=1+m1:M+m1;
if j==1+m1;
D=0; for a=-n1:n1; for b=-m1:m1;
D(n1+1+a,m1+1+b)=Lr(i+a,j+b); 
end; 
.
The noise reduction masks are presented as a normalized array to obtain a unity gain so that the noise reduction does not distort the average brightness. The figures show the result of processing a noisy image mask 1 And mask 2.
F=(1/10)*;
figure, imshow(Lvyh);
Rice. 8. Result of restoration of an image noisy with impulse noise using masks 1.
F=(1/16)*;
Lvyh=filter2(F,L,"same")/(3*3);
figure, imshow(Lvyh);
Rice. 9. Result of restoration of an image noisy with impulse noise using masks 2.
These were examples of impulse noise suppression. Let's consider similar examples of suppressing Gaussian and multiplicative noise.
Rice. 10. Result of restoration of an image noisy with Gaussian noise using masks 1 And masks 2.
Rice. 11. Result of restoration of an image noisy with multiplicative noise using masks 1 And masks 2.
Note that there are no universal methods and the processing of each image should be approached individually. If we are talking about median and low-pass filtering, then the quality of processing largely depends on good choice local aperture sizes.
The considered methods, after some modification, can be used for processing color images. Let's give an example of impulse noise suppression in a color image.
Let's take some initial image (Fig. 12):
L=imread("lily.bmp");
figure, imshow(L);
Rice. 12. Original color image.
Let's apply impulse noise with some characteristics to it:
L=imnoise(L,"salt&pepper",0.05);
figure, imshow(L);
Rice. 13. Noisy image.
For k=1:s; % processing separately for each component L=Lin(:,:,k); 
for i=1+n1:N+n1; disp(i) for j=1+m1:M+m1;
if j==1+m1; D=0; for a=-n1:n1; for b=-m1:m1; D(n1+1+a,m1+1+b)=L(i+a,j+b); end; end;
There are sad situations in life when there is not enough light, but it is not possible to open (more) the aperture or increase the shutter speed. The implication is that a “bad” photo is better than a missing photo. What should I do? Tolerate. Or use a little trick - take several frames and apply averaging.
ISO6400, it was/is now.
First, you will have to take several identical (the more, the better) photographs.
One of the series. As you can see, even at a greatly reduced size, the amount of noise is terrifying.
To average, load it all into a Photoshop document in the form of layers.
If the shooting was done handheld, the layers need to be aligned using photomerge, previously (here), or using the Edit - Auto-Align Layers command.
Next, for averaging, set the transparency of the layers: for the bottom 100%, the next 50%, 33%, 25%, ...
It is much more convenient to use the stacking mode, especially when adding large quantities photographs.
Open Photoshop and give the command File - Scripts - Load Files into Stack (File - Scripts - Load files into stack)
Check the Create Smart Object after Loading Layers checkbox, optionally Attempt to Automatically Align Source Images.
Thus we got a single group or stack. There is no need to do any magic with transparency, because... to carry out calculations on the stack there is a separate menu Layer - Smart Objects - Stack Mode (Layers - Smart object - Stack Mode). When processing photos and videos, only two modes are important - mean (averaging or arithmetic mean) and median (median), the rest are used when processing medical, scientific and forensic images, etc. If the stack mode is changed, the calculation is performed again (with the original, not the previous result).
Let's look at 100% crops from different areas and compare the results.
From left to right: original, median, averaging. A dozen frames were used.
How it works? In the case of the mean mode, the brightness of each pixel is added channel-by-channel and the result is divided by the number of photos. For example: (3+2+1+2+9+3+1)/7=3
In median mode, the average value is selected from the series 1,1,2, 2
,3,3,9 - average 2. That is strong single differences have no effect.
From a practical point of view, this means that moving objects will only leave a trace if they are present in several frames of the series. However, mean will overcome noise better.
In general, crops speak for themselves - no processing of a single frame will be able to reduce noise so effectively, because The size of the noise in this example exceeds the size of some parts.
When is this method applicable?
- when shooting with insufficient lighting at short shutter speeds (there is no shutter speed setting or it is limited by the camera, there is no tripod, there is no way to shoot for a long time, etc.)
- if necessary, reduce noise at low ISO, for example, before active post-processing.
Where won't it help?
- when shooting moving objects (although selective noise removal in stationary areas is possible).
- will not eliminate the constant component of noise
In the first part of this lesson, we looked at the reasons for the appearance of noise in photography, its components, and what to do to avoid provoking their appearance. In this tutorial we will learn how to reduce noise in Photoshop, Capture One, Digital Photo Professional And Lightroom. All of these programs have a tool for reducing noise in photography, called in photographer's jargon " noise reduction».
- On at this stage you need to understand that:
- If during shooting there are only two alternatives: to take a frame without noise (low sensitivity of the camera matrix) but blurred, or with noise but sharp, then I choose the second option. Because you can’t get rid of blurriness, but you can still fight noise.
- It is not always necessary to completely remove noise in a photograph; often it is enough to just reduce its level to an acceptable level.
- Luminance and chromatic noise are removed differently.
- At 100% scale of the image on the monitor, we see noise several times larger in area than it will be on a print, in a printed publication or online photo album.
This tutorial contains large photos that are automatically scaled if your screen size is smaller than necessary. When this happens, a button to zoom in to 100% will appear in the upper right corner of the photo. Only this scale will allow you to accurately assess the strength and size of noise. To view those parts of the illustration that are hidden, drag the image with the mouse over its central part. To close the photo and return to the article, press the Esc key.
Initial conditions: All noise reduction in my camera is turned off, filming is carried out in RAW format, the sensitivity is set to 3200 units (I still allow this value in my shooting) and 6400 units (let's see if I can use this sensitivity in an emergency). For control, a frame was shot at a matrix sensitivity of 100 units. Exposure compensation of +0.5 stops was made for all images. This slightly increased the noise level in the photographs, but exposure errors occur during shooting, so this correction is closer to the practical situation for the photographer. Test images were cut out ( photo 1): a) a fragment from Shnyr’s food packaging (to control text sharpness and color distortion); b) a scale with fields of different lightness (control of the strength of noise in different tonalities); c) a fragment of the body of the mythical creature Gava (for greater beauty). On photo 2 we see that with increasing sensitivity, noise spots grow on all target fields, which is quite natural and expected.
Photo 1: test photo.
Photo 2: increasing the sensitivity of the camera matrix leads to increased noise.
Photo 3: Reducing luminance noise reduces sharpness small parts in the picture.
How to remove noise in Photoshop.
Ah, pranksters, didn’t you shoot in RAW format or forgot to remove noise at the stage of converting the RAW file? Sometimes it happens. Open our photo in Photoshop, then go to the menu: Filter > Noise > Reduce Noise... (Filter > Noise > Reduce Noise...). Here is our first tested noise reduction.
Luminosity noise. It is the first two sliders (Strength and Preserve Details) that are responsible for reducing it. If we drag Strength to the right edge, we will see that the luminance noise is reduced, but the text also becomes blurrier ( photo 3). The main evil of brightness noise: fighting it leads to a decrease in the sharpness and detail of the photo. The attentive reader will notice that the Preserve Details slider is designed precisely to ensure that the image does not lose quality. Move the second slider further to the right and you will see the sharpness and detail return. But the noise comes back with it, so it turns out that they have exchanged the flaw for soap. Settings I used for ISO 3200: Strength – 9, Preserve Details 6%. If your photo does not have small details, such as text, texture, then Preserve Details can be reduced down to 0. For ISO 6400, these settings turned out to be weak, so I increased Strength to 10, and Details was reduced to 3%, somewhat to the detriment of text sharpness ( photo 5).
Chromatic (colored) noise seems like the lesser evil. By moving the Reduce Color Noise slider to its maximum value ( photo 4) text sharpness does not decrease, color noise almost disappears, but objects small size lose color saturation (look at the red and blue fields). Also note that a colored halo is formed around the red dies. Sometimes, such a change in the color of small details can be critical and impossible to photograph. Therefore, we should try to apply noise reduction to a minimum extent: for ISO 3200 I used a Reduce Color Noise value of 70%, and for ISO 6400 - 100%.
On photos 5 and 6 you see the result of noise reduction work in Photoshop. If for ISO 3200 after noise reduction noise manifests itself at a tolerable level and there is still some reserve for greater suppression, then for ISO 6400 they are already excessive for some shooting, and I would try in every possible way to avoid using this sensor sensitivity.
Photo 4: Reducing color noise can lead to decreased color saturation of details and color ghosting.
Photo 5: reduced noise in Photoshop, ISO 3200.
Photo 6: the result of applying Photoshop noise reduction for ISO 6400.
Conclusions: Reducing luminance noise is impossible without reducing the sharpness of the photo. The use of noise reduction makes it possible to use a sensitivity of 3200 units, but a sensitivity of 6400 may not be suitable for increased requirements for photographic quality. If you take photographs for the Internet or small prints, then I can use a sensitivity of 6400 units. By reducing brightness noise in photography, we do not get rid of chromatic noise, and vice versa.
Reducing chromatic noise in photography can sometimes go unnoticed by the viewer. But if when shooting, color accuracy in small details is important, then excessive use of noise reduction settings in Photoshop is unacceptable, for example when subject photography or in food photography. The more “gentle” the noise reduction settings we use (not only in Photoshop, but in general), the better quality our image is after processing.
Digital Photo Professional
The second one for this lesson I chose Canon Digital Photo Professional(hereinafter referred to as DPP). This is a very simple RAW file converter for Canon cameras and it is with its help that I introduce students in a photography course for beginners to the capabilities of the RAW format. In order to get to the DPP noise reduction, you need to select the NR/Lens/AOL tab on the Tool Palette. Naturally, we are interested in the Noise reduction block, which contains only two sliders: Luminance... - for reducing brightness noise, and Chrominance... - for chromatic ( photo 7). As with Photoshop's noise reduction, I tried to apply the following settings in DPP to maintain a balance of quality for small details and smooth surfaces. For ISO 3200 the following parameters were used: Luminance - 7, Chrominance - 12 ( photo 8). For ISO 6400 - 12 and 20 respectively ( photo 9). The result is very similar to the one obtained in Photoshop's noise reduction.
Setting up noise reduction in DPP. I noticed that when my camera's noise reduction is turned off, DPP applies its own noise reduction to RAW files. It is not convenient to turn off photo noise reduction every time, so you need to make sure that DPP does not apply it by default. To do this, go to the DPP settings (Ctrl + K keys), go to the Tool palette tab, turn on the Set as defaults switch, set all sliders to 0, click OK, and restart DPP ( photo 10).
Photo 7: Canon Digital Photo Professional noise reduction.
Photo 8: the result of applying DPP noise reduction for ISO 3200.
Photo 9: the result of applying the same noise reduction for ISO 6400.
Photo 10: DPP noise reduction settings.
Capture One
Today Capture One is my main RAW file converter. As with DPP, its noise reduction ( photo 11) is not disabled, and is applied to the RAW file regardless of camera settings. Moreover, even when there is no need to reduce noise, for example, with low sensitivity. I did a little research on the noise reduction algorithm in Capture One, and it interested me so much that I decided to read the help of this RAW converter. Alas, none useful information I couldn’t find any information on how noise reduction works in Capture One. Therefore, the results of my conjectures, assumptions and research will be described below.
According to the Capture One help, the noise reduction of this converter changes its settings after analyzing the file. I admit, over several years of working in Capture One, I have adjusted its noise reduction settings only a few times. Noise reduction works so gently, intelligently, unobtrusively and excellently in automatic mode that I simply forgot about its existence.
The first thing I checked was how my photography would improve when I removed the noise reduction settings for ISO 100. And nothing happened. That is, if there is no noise, then the noise reduction does not work. Then I noticed that increasing the sensitivity only changed the Color value (the effect on color noise), but not the Luminance value (brightness noise). Then I assumed that with the same Luminance value and with increasing sensitivity, luminance noise would increase in proportion to what happens in the absence of noise reduction. Not so. The noise increased, but not so significantly. I won’t guess how this happens, but I was pleased with the result of Capture One’s intelligence.
In the following experiment, I tried to find the minimum value of the noise reduction settings that would satisfy me, and compare how much softer my settings were with the default settings offered by Capture One. The changes were so minor that they can be ignored: for ISO 3200 Capture One suggested values of 25 and 54 (Luminance and Color), but I found softer values acceptable: 20 and 50, respectively. For ISO 6400, Capture One's own noise reduction settings completely satisfied me, and I did not touch them (25 and 57).
There are a few more amenities that can make noise cancellation even more effective. Surface reduces large noise spots on low-contrast, smooth surfaces without affecting fine details such as text (value 70 for ISO 3200 and 90 for ISO 6400). Single Pixel allows you to remove single-pixel noise (individual knocked-out pixels) without losing fine details. True, such pixels appear only at ISO 6400 or when the matrix overheats in Live View mode. Despite the fact that the noise reduction test used a matrix sensitivity of 6400 units, I did not use this Capture one setting, since the impact of the main tools was sufficient.
I'm very pleased with the quality and noise reduction capabilities of Capture One. Unlike the noise reduction devices discussed above, Capture One does not create color halos or reduce color saturation in small details of the photo. Colored noise in shadowed areas is also suppressed significantly better than previous competitors. This indicates the high quality of the color noise reduction algorithm. Surface's action also makes luminance noise appear weaker, especially on plain surfaces.
You can see the results of noise reduction in Capture One at photos 12 and 13. However, it remains to test a competitor among RAW converters - noise reduction in Lightroom.
Photo 11: Capture One noise reduction.
Photo 12: the result of using Capture One noise reduction for ISO 3200.
Photo 13: the result of applying the same noise reduction for ISO 6400.
Lightroom and Adobe Camera RAW
I even downloaded the new Lightroom - 4.3... In all previous versions of Lightroom, its noise reduction, according to users, was considered a weak link and was not recommended for use. Those. After converting the RAW files into Lightroom, noise reduction had to be done in Photoshop. But Photoshop's noise reduction is much inferior in quality to at least Capture One, and I can't recommend it at all this chain(Lightroom > Photoshop) for noise reduction. I have come across a mention on the forums that Lightroom noise reduction has been improved, starting with the fourth version. Wanting to clarify this information with experienced users, I again found myself on the Lightroom forums. And what I read there did not please me at all: slowdowns, difficulties in operation, glitches, in general, everything is as always with the RAW file converter from Adobe. This finally turned me away from installing Lightroom, and instead of its noise reduction, I will test a similar Photoshop tool - Adobe Camera RAW. I have long noticed that the settings of these two Adobe products are identical, and lead to the same results after processing RAW files. That is, the operating algorithms of both programs are the same (it would be strange for one manufacturer to make two versions of noise reduction). If I'm wrong and you have good reasons for this, please let me know.
In order to reduce noise in Adobe Camera RAW, you need to go to the Detail tab. This noise reduction has more settings than the noise reduction in Photoshop (photo 14). By default, for files of both sensitivities, Camera RAW suggests not reducing luminance noise, but reducing color noise (Luminance – 0, Color – 25, Color Detail – 50). At these settings, color noise is suppressed beautifully, and (as in Capture One) I don't notice any color ghosting. Wonderful. The Color Detail slider helps you adjust (return) color saturation for small parts(remember, there was a problem with this in Photoshop noise reduction). I left the Color Detail value as default, i.e. 50. But I lowered the main Color setting to 15 (for ISO 3200) and 20 (ISO 6400).
Highlight spots on smooth surfaces remained visible but unobtrusive at Luminance settings of 55 (at ISO 3200) and 70 (ISO 6400), but did result in a slight reduction in text detail. Therefore, I chose a compromise Luminance Detail value of 40 (for ISO 3200) and 50 (ISO 6400).
I really liked how Adobe Camera RAW reduces noise ( photos 15 and 16) that I thought about the permissibility of wider use of sensitivity 6400 on my camera. If we added the Surface setting to this noise reduction as in Capture One, then it would have no equal. I wonder how the places will be distributed among the noise reduction leaders at the end of this photography lesson.
Photo 14: Adobe Camera RAW noise reduction (the settings are identical to Lightroom).
Photo 15: the result of applying Adobe Camera RAW noise reduction for ISO 3200.
Photo 16: the result of applying the same noise reduction for ISO 6400.
Noise reduction test results
Noise reduction test results in photos 17 and 18: the worst are at the top, the best are at the bottom. When shooting at high sensitivities, I do not recommend using Photoshop noise reduction for raster images and Canon Digital Photo Professional. The main reason is strong color halos around colored parts in a photograph. It is also difficult to find a compromise in these noise cancelers between the level of brightness noise on smooth surfaces and the sharpness of small details. Capture One, compared to the first two, looks in an advantageous position until the Adobe Camera RAW noise reduction comes into play. The latter showed that in many cases I could use 6400 sensitivity for commercial shooting as well: amazing luminance noise reduction for smooth surfaces while maintaining fine details and good job to reduce color noise. I don’t understand what people who work in Lightroom are complaining about?
Photo 17: comparison table noise reduction for ISO 3200.
Photo 18: Comparison table of noise reduction for ISO 6400.
Photo 19: Charming noise.
Conclusion
If you want to use high sensitivity values when shooting, then start fighting noise at the shooting stage - shoot in RAW format. Do not transfer the process of noise reduction to Photoshop; do it in a RAW converter when correcting photos. Use the converter that will reduce noise in photography with less losses (and losses are inevitable). Use minimum values noise reduction installations.
If the photo has loud noises, then in some cases you can limit yourself to reducing only the color spots. The remaining luminance noise will be very similar to film grain. Sometimes this grain imitation is even preferable to the smooth image of a digital camera. For example, if you are stylizing a photograph as antique. In other cases, grain can give a certain charm to a photo ( photo 19). It’s not for nothing that there are Photoshop filters that create similar film grain. But, this is a topic for another lesson.
PS: This noise reduction test did not use the latest versions of Adobe Camera RAW and Capture One. Therefore, it is possible that the noise reduction algorithms in these programs have become even more advanced.
PPS: Don't make noise!
Loading...