Grid Enabling a Content Based Image Retrieval Application

摘要

Image clustering is a vital step in Content Based Image Retrieval (CBIR). One popular algorithm for image clustering is based on genetic algorithms, a widely used technique for search and optimization problems and belongs to the group of evolutionary algorithms. Serial implementations of the algorithm, customized for image clustering, suffer from slow execution rate. Also, a genetic algorithm, if run for small number of generations, gets stuck in the local minima, and therefore requires larger number of generations to achieve optimal results. These two factors motivated the parallel implementation of the algorithm. The concurrency in the application was identified and partitioned such that the I/O and computation were done in parallel. The application was implemented using "Master-Worker" paradigm and "All-Worker" paradigm in C and MPI. Different load balancing schemes were also tested and resulted in a linear speedup and vastly reduced the execution time. The implementations were further modified to run successfully in Grid Environment where resources are heterogeneous, distributed across disparate locations, and can appear and disappear dynamically. This paper demonstrates that performance of CBIR applications can be noticeably improved if the genetic algorithm used for clustering images is run for larger number of generations. The benefits of check-pointing and restart mechanisms in grid applications are also demonstrated.