Concurrent ML as an Alternative Parallel Programming Style for Image Processing

Many approaches to simplifying/enabling the use of parallelism in programming tasks exist. A substantial number of those concentrate solely either on data-parallelism or fork-join-based task-parallelism. Concurrent ML is an alternative approach based around the concept of lightweight individual sub-processes synchronously exchanging messages. This work explores the usefulness of the CML approach in image processing, by applying it to the basic median filter operation and contrasting it with other simple implementations. The results strongly suggest that it is a comparatively poor fit to such an operation, with the one slight advantage of apparently having better peak memory requirements. It is not clear, however, how efficient are either the algorithm implementation or the library it is based on.