Algorithm for MEDIAN Filter on Parallel Processor
Original Publication Date: 1988-Aug-01
Included in the Prior Art Database: 2005-Feb-15
An algorithm is disclosed for a median filter on a single instruction multiple data (SIMD) parallel or array processor for use in image processing applications. This enables a superset of the median filters, e.g., MAX, MIN, MEDIAN to be efficiently implemented which would otherwise be extremely inefficient. A minimum amount of data is transferred at each step. A typical application is where an image consists of 1024 x 1024 8-bit pixels. For each pixel it is required to find the median value of that pixel and its 8 nearest neighbors. An obvious method is to consider each pixel in turn, together with its 8 nearest neighbors, sort these 9 into ascending order and pick the middle (fifth smallest) value. This requires a sort, which is difficult to implement on SIMD processors, and slow on any processor.