[HN Gopher] Median filtering (2022)
       ___________________________________________________________________
        
       Median filtering (2022)
        
       Author : todsacerdoti
       Score  : 70 points
       Date   : 2024-07-23 12:17 UTC (10 hours ago)
        
 (HTM) web link (www.crisluengo.net)
 (TXT) w3m dump (www.crisluengo.net)
        
       | foooorsyth wrote:
       | Had a lot of fun implementing the Huang histogram approach and
       | the constant time approach by hand in a prior role.
       | 
       | Never saw the binary tree approach. And this article, being
       | written in summer of 22, missed out on the 2D wavelet approach
       | published later on that year.
       | 
       | https://cgenglab.github.io/en/publication/sigga22_wmatrix_me....
        
         | tomrod wrote:
         | I'm interested to learn more about these algorithms. Are there
         | any sources you'd recommend?
        
           | teruakohatu wrote:
           | Not the OP but have a look at the scikit-images and other
           | scikit packages. Lots of different algorithms implemented.
           | Also of course the OpenCV library.
        
       | hnmullany wrote:
       | I hacked together an SVG-filter based median filter once. It was
       | horribly inefficient, but it worked:
       | https://codepen.io/mullany/pen/ngJWvx
        
       | __abadams__ wrote:
       | I've found sorting-network-based filters to be faster than these
       | methods up to size 25x25 or so. It has worse computational
       | complexity, but filters up to that size covers a lot of ground in
       | practice. See Figure 10 in
       | https://andrew.adams.pub/fast_median_filters.pdf
        
       | ttoinou wrote:
       | Median filtering with large kernel can be pretty ugly as it gives
       | pixels who are far away the same weight. I prefer weighting them
       | (for example with a gaussian) and using a weighted histogram, use
       | the weighted median value (interpolated between two values). It
       | gives a result even more interesting (for some uses) than a
       | simple gaussian blur and you can also configure the weights to
       | make it edge preserving like a bilateral blur but better
       | There is also no way to split up the median computation
       | 
       | What does this mean here ? Seems like we could have a rolling
       | window by adding and subtracting pixels on the way. I've coded
       | this before, although it's not O(1) like the algorithm described
       | at the end
        
       | AcerbicZero wrote:
       | For a second, I thought median filtering was going to be about
       | riding motorcycles down the median of a road to filter through
       | traffic.....
        
       | infocollector wrote:
       | Does anyone know of a pytorch differentiable filter that does the
       | same thing?
        
       ___________________________________________________________________
       (page generated 2024-07-23 23:04 UTC)