Image Data Mining Based on Parallel Association Rules
|School||Liaoning Technical University|
|Course||Applied Computer Technology|
|Keywords||Association rules Parallel Computing Fleet Frequent Itemsets Candidate itemsets|
This paper studies the multimedia data mining prototype , starting from the architecture established containing media library feature library , repository of multimedia databases , in order to fully demonstrate the characteristics of the image data , thereby solving the problem of the representation of the image data itself . Meanwhile, to improve mining efficiency , the use of high-performance parallel cluster environment , designed to improve the parallel association rules algorithm for efficient parallel excavation . Unified messaging using MPI average allocation of resources in order to achieve higher load balancing efficiency ; using master -slave architecture , a processor as the main control processor dedicated to generate the global frequent itemsets , and is responsible for the information and other processors interaction , while other processors as a slave processor, the only local candidate is responsible for generating sets and counting and pruning, there is no information from the interaction between processors , so that you can reduce the communication time and improve efficiency ; in establishing a master controller Hash table to achieve frequent itemsets in the database table records the mapping to improve the speed of queries and counting . The experiment proved that the proposed algorithm improvement and optimization of the serial association rules algorithm, can guarantee candidate itemsets and frequent itemsets integrity: the classical parallel algorithm of association rules CaD and CD , compared with a higher speedup and efficiency.