MPIIMGEN - a code transformer that parallelizes image processing codes to run on a cluster of workstations

An enormous body of image and video processing software has been written for conventional (sequential) desktop computers. These implement a wide range of operations, such as convolution, histogram equalization and template matching. These applications usually have a tremendous potential for parallelism. However a significant barrier in exploiting such parallelism is the difficulty of writing parallel software. In this work, the design and implementation of MPIIMGEN -- a code transformer that automatically transforms these sequential image processing codes into parallel codes that are capable of running on a cluster of workstations is presented. This tool uses a pattern driven approach to parallelize the sequential codes.