Parallel implementation of vision algorithms on distributed systems

Vision computing involves the execution of a large number of operations on large sets of structured data. The need to implement vision tasks in parallel arises from the speed requirements of real-time environments in various application domains. In this paper we propose that a distributed computer system can be utilized to replace the specialized machine for the parallel implementation of vision tasks. We introduce some techniques used in distributed systems and adopt a divide-and-conquer policy to schedule the complex vision tasks for parallelism. Two traditional vision algorithms for matrix operation and image matching are implemented using PVM (parallel virtual machine). Furthermore, a hierarchical object recognition system is described as an example of parallelism on distributed systems. Finally we conclude that some vision tasks can be realized on a general distributed system to achieve the speedup at a low cost.