Fast Computing Median Filters On General-Purpose Image Processing Systems

Abstract
Separable median filters over 3 X3 neighborhoods can be realized by applying in succession logical operations in the horizontal and vertical directions. We show how to take advantage of the architectural resources of commercially available pipelined image processing systems in implementing these filters. The required logical operations are decomposed into a succession of image additions, subtractions, and translations that can be executed very efficiently. The proposed method showed a 70-fold improvement of processing speed over the well-known histogram updating method for 3 X3 filters over 512 X512 X8 bit images.

This publication has 0 references indexed in Scilit: