Algorithm performance indexing through parallelism

Image processing of Spotlight Synthetic Aperture Radar (SSAR) involves large amount of data to be processed and the time required to handle such data is relatively large. Evaluation of an algorithm in terms of processing speed (run time) and space (data structures) is therefore a critical step in ascertaining suitability of an algorithm for a particular application. Major limitation in physical processing speed is principally associated with the complex computations required to provide solutions to large size problems (Jaja, 1992). With an increase in massive parallelism of processors, the overall processor operational speed has drastically improved especially in image processing. The best choice of an algorithm on a particular processor and for a particular task is based on the understanding of the performance of the algorithm and hardware at hand. In this paper, we address this problem by providing a means to know the characteristics of a given algorithm by evaluating its performance. Two segmentation based SSAR algorithms, namely the 1-D and 2-D data segmentation are evaluated and algorithm performance indexes drawn. In both cases, image data was split into segments and parallel processed by individual asynchronous processors. Based on the relationship between the non-overlapping data segment size and execution time, the performance index of an algorithm was determined.