PAMIR: Parallel Multimedia Information Retrieval System

This paper describes the design, implementation and evaluation of Parallel Multimedia Information Retrieval systems. PAMIR is implemented on a network of Pentium PC based workstations to exploit the I/O and computation parallelism in both insertion and processing phase of complex similarity queries in multimedia Information retrieval. The system is capable to index and query multimedia objects concurrently based on more than one features (i.e., global color, local color, texture, shape) and each features of an object queried by a different workstation using multithreaded version of M-tree. The system was tested for a collection of approximately 68000 images and each image described by three features; color moments, histogram layouts and texture. Experimentation has been shown that the system provides a good reduction in CPU times on both insertion and query processing phase. Detailed explanation of the system and experimentation results also presented.