Generic Filter (Majority)

With the Spatial Generic Filter Processing Algorithm you can use the scipy generic_filter function to write your own spatial filter functions.

In this example we will write a mode filter (also known as majority filter). This kind of filter is often used to smooth classification results (reduce salt-and-pepper effects in the image). It calculates the mode of all values in a specifiable neighborhood around a pixel (e.g. 3x3 pixel window) and assigns this value to the pixel.

  1. In the Processing Toolbox go to and open Convolution, Morphology and Filtering ‣ Spatial Generic Filter

  2. Select an input raster under Raster

  3. In the Code text window you can enter python code. Delete the existing content and enter the following

    from scipy.ndimage.filters import generic_filter
import numpy as np
from scipy.stats import mode

def filter_function(invalues):
   invalues_mode = mode(invalues, axis=None, nan_policy='omit')
   return invalues_mode[0]

function = lambda array: generic_filter(array, function=filter_function, size=3)

  4. Specify the output raster location and click Run

../../_images/modefilter.png

Majority filter applied to classification image: original classification (left), majority filtered 3x3 window (middle), majority filtered 5x5 window (right)

Tip

Also have a look at the scipy.stats.mode documentation. You can change the window size by altering the size parameter in the generic_filter function.

You could further improve the function above by putting constraints on the definition of majority (for example, only update the original value if the frequency of the modal value is higher than 50 percent)