Cris Luengo explains the concept of the median filter, a non-linear smoothing filter that calculates the median value of a neighborhood in an input image. The filter can be generalized into a percentile and rank filter, useful for noisy images. Various algorithms, like histogram-based and binary tree, are discussed for efficient median computations. The O(1) algorithm by Perreault and Hérbert significantly speeds up the process for 8-bit images and square kernels. DIPlib implementation showcases improved performance with single-precision floating-point pixels. Comparisons show different implementations and efficiencies in processing times based on the number of values in the input image.
https://www.crisluengo.net/archives/1138/