Trapping noise with Gaussian wavelets

    (by Cor Berrevoets )


Sometimes you find that when using wavelets you get to see more noise instead of details when enhancing the image. This noise is often most visible in wavelet-layer 1 but can often also persist in other layers. RegiStax wavelets allow you to "trap" part of the noise so the effects of the noise do not show up in the higher layers.

Using Gaussian
You will need to use the Gaussian wavelets as they allow more control of each wavelet-layer. The gaussian filters are most of the time (also when you do not have much noise) the best way to enhance image details with most user control.

First make sure all the Gaussian filters are set at 0.1 (the default). Then uncheck layer 1 (this will be the layer we are trying to trap the noise into) or set the slider of layer 1 to 0.0). Unchecking a layer makes the details in that layer invisible. 

Now move the slider of layer 2 upward until you see the noise appearing in the image. You might even try to reduce the filter-size of layer 2 but thats not necessary (figure 1). 

When the noise is clearly visible we will try to trap the noise into layer 1 by enlarging the gaussian filter of layer 1. Just increase the gaussian filtersize using the up/down control next to the filtervalue. Do this step by step until you see the noise gradually disappearing. The figures below (figure 2 and 3) show the effect of simply increasing the filtersize of layer 1. Be aware that as a consequence of increasing the filtersize of layer 1 you might loose details. Figure 3 in the example below clearly shows less details than Figure 2. As always every stacking-result will demand carefull tuning of wavelets to get the most out of your images. The settings I have used in the example below are therefore not by any chance "golden settings". You will need to experiment yourself to find out what kind of wavelet-settings your images need.

Cor Berrevoets

Figure 1: visible noise in the image


Figure 2: After deselecting layer 1 and increasing the filtersize of layer 1 from 0.10 to 0.15


Figure 3: After increasing the filtersize of layer 1 from 0.15 to 0.20


Cor Berrevoets