File Exchange

## Min/Max filter

version 1.7 (28.7 KB) by

Multidimensional non-linear min/max filtering

Updated

The filter computes the minima and/or maxima of an array over sliding window with a given size.

Multidimensional array is fully supported: running filter in 1D, 2D filter for image processing applications (erosion/dilatation), 3D and more.

This package has been implemented with a special care on the running speed: the MEX engine uses an algorithms that requires no more than three (3) comparisons per element and per dimension in all configurations.

All numerical and logical class arrays are supported.

Contributor (beside author) is Vaclav Potesil

Bruno Luong

### Bruno Luong (view profile)

To Adhithya: The algorithm reference is cited inside the source code LEMIRE_ENGINE.C

I am using your minmax filter for 3D data. Do you have any flowchart to understand how the long code works for min/max filter.

Is it possible to only get the summit locations?

Chris Volpe

### Chris Volpe (view profile)

Stupid question: How is this different from simply doing an erosion (min) or dilation (max)? Or is this just for people who don't have the image processing toolbox?

ali

seongjin

### seongjin (view profile)

So far, this is the fastest that I know. Thank you.

Wathsala Ranasinghe

Gerald Grabner

Fast!

Qiao Liyong

shi

Michael Teo

### Michael Teo (view profile)

Thanks for sharing, its make sense and it solved my problem. :)

Bruno Luong

### Bruno Luong (view profile)

I copy the question from Michael Teo: Does it handle NaN data?

<QUOTE>
ePeriod = 3;
eData = [ 5;1;3;NaN;8;2;3;NaN;1;9 ];
minmaxfilt(eData, ePeriod, 'max', 'valid');

Actual output:
[5;3;3;NaN;8;3;3]

If we take NaN as a empty data, the expected output is:
[5;3;8;8;8;3;3]
</ENDQUOTE>

The answer is "no". The work around id to replace NaN by -Inf for running max and +Inf for running min.

eData(isnan(eData))=-inf
minmaxfilt(eData, ePeriod, 'max', 'valid')

Tobias

### Tobias (view profile)

Nice Max-Filter. Works faster than implementation of Frederico D'Almeida and seems to be correct.