Grid Enabling a Content Based Image Retrieval Application

摘要

Image clustering is a vital step in Content BasedImage Retrieval (CBIR). One popular algorithm forimage clustering is based on genetic algorithms, a widelyused technique for search and optimization problems andbelongs to the group of evolutionary algorithms. Serialimplementations of the algorithm, customized for imageclustering, suffer from slow execution rate. Also, agenetic algorithm, if run for small number of generations,gets stuck in the local minima, and therefore requireslarger number of generations to achieve optimal results.These two factors motivated the parallel implementationof the algorithm. The concurrency in the application wasidentified and partitioned such that the I/O andcomputation were done in parallel. The application wasimplemented using "Master-Worker" paradigm and "All-Worker" paradigm in C and MPI. Different loadbalancing schemes were also tested and resulted in alinear speedup and vastly reduced the execution time. Theimplementations were further modified to runsuccessfully in Grid Environment where resources areheterogeneous, distributed across disparate locations, andcan appear and disappear dynamically. This paperdemonstrates that performance of CBIR applications canbe noticeably improved if the genetic algorithm used forclustering images is run for larger number of generations.The benefits of check-pointing and restart mechanisms ingrid applications are also demonstrated.