4. Denoising & Smoothening

<< Back to Index

4.1 Denoising

The denoising menu is used to reduce or remove the noise in the image the may have become very visible after applying sharpening.

4.1.1 Bilateral

This is the preferred way to remove noise. This algorithm is one of the most effective denoising algorithms. It is very effective in removing noise pixels (de-speckling) while also preserving sharp edges. Be aware though that an unwanted effect may be that certain edges may start to show the so called "saw effect". This effect can be compensated by combining denoising with smoothening (see 4.2). Bilateral denoising has the following parameters:

Sigma color

This parameters represents the intensity of the effect applied.

Radius

This is the radius area expressed in pixels to which the algorithm is applied. It is recommended to keep this value set to either 1 or 2. Higher values will also slow down the processing.

Iterations

This parameters control the number of times that the algorithm is applied. In particular for Bilateral denoising it is recommended to use multiple iterations. The effectiveness of the algorithm is improved. In general this will improve the edge preservation nature of the algorithm while at the same time being very effective in removing the noise pixels. Make sure to use a low value for Sigma color when applying multiple iterations.

4.1.2 Sigma 1 denoising

Sigma denoising is the other algorithm supported. Sigma 1 is somewhat less effective than Ian's NR, but is a good choice for images with few or very fine noise, and is very effective at preserving details versus noise patches (outliers). In case you did not install G'MIC (only applies to non windows versions) Ian's NR will not be available. In that case sigma 1 denoising is a good alternative. The filter can be controlled using the following sliders:

1 Amount: Higher values will increase the noise reduction and is better as preserving edges but will also be less effective on noisy images.

2. Radius: Allows to select the pixel size where the filter is applied to. The larger the radius chosen the more pixels it will use to average out the noise.

3. Iterations: this is the number of times that the Sigma filter algorithm is applied.

4.1.3 Ian's NR

This filter has very effective de-speckling capabilities, similar to Bilateral. It will remove noise while at the same time keeping almost all details in tact. The animation on the right shows an example. This filter is from the well known open source G'MIC image processing tool. In LuckyStackWorker the filter can be controlled by 4 sliders: Fine details, medium details, Recovery and iterations. By inceasing the fine or medium details a stronger denoising will be applied. The fine details will work on the finest speckles while medium will remove larger speckles and artefacts.

The recovery is meant for recovering details that might have been wiped out by the application of the filter.

Finally iterations will simple apply the algorithm multiple times. In some cases this can be more effective then applying a one time Ian's denoising with very high values for the fine or medium details.

Note that Ian's noise reduction is applied by the G'MIC library which tends to be rather slow. Especially on larger images this may slow down the overal experienced processing quite a bit. For faster processing I recommend using one of the native denoising algorithms such as Bilateral or Sigma 1.

4.2 Smoothening

Smoothening is meant to obtain a smoother and more natural outcome and the choice of the 2 filters are very effective at this. Denoising algorithms such as Bileteral (see 4.1) may result in the so called saw-effects or edges that look unnatural. By applying smoothening these effects can be reduced.

4.2.1 Savitzky-Golay

This option is based on the well known and very effective 2D Savitzky-Golay smoothening algorithm. This algorithm is very effective in obtaing a smoother result while at the same time preserving the contrast. It can be controlled with 3 sliders. The Kernel Size slider selects the algorithms kernel size. A smaller size will work better on very fine noise but is less strong than when you choose a larger kernel size. Larger kernel sizes work better on larger noise patches, but may also blur out fine details. The amount slider can be used to blend the non denoised version of the image with the fully denoised version based on the choose kernel size. Finally the number of iterations stands for the number of times that the algorithm is applied. Applying multiple iterations can result in better detail preservations and less blurring out of fine details while still applying a smaller kernel size.

4.2.2 Sigma 2

Sigma2 is the other algorithm supported. Although it can obtain very similar results as the Savitzky-Golay algorithm, the latter is better at preserving contrast. The Sigma filter 2 algorithm uses 2 sliders. Sigma mode 2 behaves more like a traditional averaging filter. Applying this will result in more noise reduction but it will also soften edges. The following parameters can be adjusted:

1. Radius: Allows to select the pixel size where the filter is applied to. The larger the radius chosen the more pixels it will use to average out the noise.

2. Iterations: this is the number of times that the Sigma filter algorithm is applied.

Denoise channels individually

As of version 6 it is possible to apply denoising to the red, green or blue channel individually. Note that this feature only works when Sigma 1 denoising is selected for Denoise pass 1 (it cannot be used in conjunction with Ian's noise reduction).

To do so, select the desired channel to process by clicking one of the radio buttons "All", "Red", "Green" or "Blue" on the "Apply to" selector on the bottom of the denoising tab. The image panel will automatically switch to the visible channel. To edit another channel simply click the desired channel.

To show the end result of all channels in color you can either click the channel switcher on top on the control panel until "RGB" is shown, or click the "All" option. Be aware that when switching to the "All" option means that you will apply all any follow up changes to all channels again, so making any changes to the denoising sliders when "All" is selected will overrule any custom changes made to the individual channels.

Google Sites

Report abuse