Mode filter

This smoothing filter, sometimes called ‘majority’, selects the mode average of the pixels in the support. It is the best way to filter discrete attributes (those that take only certain values). The most common example of this type of attribute is waveform class, which typically takes integer values from 1 up to 10 or so (see Figure 4 in Hall 2007). Smoothing an attribute like this with, say, a mean filter would result in lots of undefined values halfway between two classes. In contrast, the mode filter does a good job of smoothing this type of data while preserving the integers. In my experience, it is quite aggressive, and supports larger than 7 × 7 are rarely called for.

Naturally, for a continuous variable (like time or amplitude), there is unlikely to be a majority of any one value in the support, so this filter is not suitable for these attributes; indeed, it is prone to artifacts in such data.

See also

• Smoothing filters
• Median filter — a useful conditioning step