# Perona Malik Noise Estimator using Canny

**URL:** <https://discourse.itk.org/t/perona-malik-noise-estimator-using-canny/384>\
**Category:** Algorithms\
**Created:** [November 2, 2017, 2:03pm UTC](https://discourse.itk.org/t/perona-malik-noise-estimator-using-canny/384 "2017-11-02T14:03:02Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![AlexRothberg](https://discourse.itk.org/letter_avatar_proxy/v4/letter/a/e9a140/32.png) [@AlexRothberg](https://discourse.itk.org/u/AlexRothberg)\
**Post date:** [November 2, 2017, 2:03pm UTC](https://discourse.itk.org/t/perona-malik-noise-estimator-using-canny/384/1 "2017-11-02T14:03:04Z")

</div>

In [Perona and Malik](http://bioen.utah.edu/wiki/images/d/d8/peronamalik90.pdf), they state:

> a histogram of the absolute values of the gradient throughout the  
> image was computed, and `K` was set equal to the `90%` value of its  
> integral

My interpretation of this is to take the 90%ile of the distribution of magnitude of gradients. This interpretation however differs from the implementation [here](https://github.com/CSIM-Toolkits/ITK/blob/d8c89499eedeb228f5b127e11055bc245deb891a/Modules/Features/AutomaticConductanceImageCalculator/itkAutomaticConductanceImageCalculator.hxx#L119-L132) which seems to do a double cumulative sum.