Smooth

This method allows you to reduce noise in an image. Each pixel is replaced by a weighted average of its neighbors. The weighting factors are calculated from the binomial coefficients in accordance with the filter size. The binomial filter is very similar to a Gaussian filter in its effect.

This method removes noise from images using wavelet transformations or total variation. The process of denoising an image with wavelet transformations can be broken down into the following three parts:

• Calculate the wavelet transform of the noisy image.
  The wavelet transformation can be calculated by the method Real Wavelets and Complex Wavelets.
• Modify the noisy wavelet coefficients.
  This is done by using bivariate shrinkage with local variance estimation (thresholding). [Bivariate Shrinkage with Local Variance Estimator, Levent Sendur and Ivan W. Selesnick, IEEE Signal Processing Letters, Vol. 9, No. 12, December 2002]
• Compute the inverse transform using the threshold coefficients.

This method allows you to reduce noise in an image. Each pixel is replaced by a weighted average of its neighbors. The neighboring pixels are weighted in accordance with a two-dimensional Gauss bell curve.

This method allows you to reduce noise in an image. Each pixel is replaced by the average of its neighbors. The size of the area of the neighboring pixels considered is defined by a quadratic filter matrix. The modified pixel is the central pixel of the filter matrix.

This method allows you to reduce noise in an image. Each pixel is replaced by the median of its neighbors. The size of the area of the neighboring pixels considered is defined by a quadratic filter matrix. The modified pixel is the central pixel of the filter matrix. The median is the middle value of the gray values of the pixel and its neighbors sorted in ascending order.

This method allows you to reduce noise in an image. Each pixel is replaced by the average of its neighbors. The size of the area of the neighboring pixels considered is defined by a quadratic filter matrix. The modified pixel is the central pixel of the filter matrix. To calculate the average, only the gray values that lie within a defined range (+/- sigma) around the gray value of the central pixel are taken into consideration. As a result, fine object structures are not blurred; only the gray levels in image regions that belong together are adjusted.

With this function you can remove single pixel phenomena, such as those that occur in the case of clocking induced charge with EMCCDs and as radio telegraph signal noise with CMOS sensors. It is a filter which analyzes the input image and removes pixels, whose intensity value diverges strongly from the median intensity of its neighboring pixels.

The filter analyzes the input image and removes pixels that are "much" larger than the median of their neighbors. The algorithm works as follows:

1. Sort all 9 pixels in a 3 x 3 neighborhood.
2. Determine the median intensity value of the sorted pixels.
3. Multiply the median by the threshold factor to get the limit.
4. If the center pixel intensity is larger than the limit, replace the pixel with the median.

Larger values of the threshold factor increase the value of the intensity limit and decrease the number of pixels that are replaced. The default value of 1.5 is arbitrary, but seems to remove charge induced noise from images acquired using cameras with EM gain capabilities. This filter can also be used to remove hot pixels from images.

This method removes "correlated noise components" from an image, resulting in an image with so-called "white noise", where the noise in every pixel is not correlated with the noise found in neighboring pixels. Such an ideal noise is also called "White Noise". This function should be used to pre-process image data that is later used to train an Noise2Void denoising model using the ZEN Intellesis Denoising. Keep in mind that applying a model trained on "whitened" datasets should be only applied to "whitened" images.