Image Filtering: Difference between revisions

Created page with "Filtering an image refers to sampling an image in a way that resolve aliasing. Since images are a 2D signal, image filtering is a type of [[Wikipedia: Filter (signal processing) signal filtering]. ==Denoising== ===Mean filter=== Simply take the mean of all pixels in the neighborhood. ===Median filter=== Median of pixels in a neighborhood. This preserves edges. ===Gaussian filter=== Convolve the image with a gaussian kernel.<br> This is similar to a mean filter but p..."
 
 
(2 intermediate revisions by the same user not shown)
Line 1: Line 1:
Filtering an image refers to sampling an image in a way that resolve aliasing.  
Filtering an image refers to sampling an image in a way that resolve aliasing.  
Since images are a 2D signal, image filtering is a type of [[Wikipedia: Filter (signal processing) signal filtering].
Since images are a 2D signal, image filtering is a type of [[Wikipedia: Filter (signal processing) | signal filtering]].


==Denoising==
==Denoising==


===Mean filter===
===Mean filter===
Simply take the mean of all pixels in the neighborhood.
Simply take the mean of all pixels in the neighborhood. Also known as a box filter.


===Median filter===
===Median filter===
Line 17: Line 17:
{{ main | Wikipedia: Bilateral filtering }}
{{ main | Wikipedia: Bilateral filtering }}
A bilateral filter is a gaussian filter but pixels are additionally weighted by the gaussian of the intensity difference. Hence, edges are preserved since adjacent pixels which have significantly different intensity are weighted much less.
A bilateral filter is a gaussian filter but pixels are additionally weighted by the gaussian of the intensity difference. Hence, edges are preserved since adjacent pixels which have significantly different intensity are weighted much less.
This can be used for upsampling lower resolution depth maps or tone maps using [https://johanneskopf.de/publications/jbu/paper/FinalPaper_0185.pdf joint bilateral upscaling]. E.g. if you have a low-resolution image, low-resolution depth, and a high-resolution reference, you can upscale the depth while keeping the edges from the high-resolution reference.


==Upfiltering==
==Upfiltering==