Parallel Implementation of Ant Colony Optimization for Vector Quantization Codebook Design

This paper presents a framework for parallel implementation of ant colony system-based vector quantization codebook design. The most important structure used is the pheromone trail, which is updated in both local and global sense. The local renewal is implemented in each processor and the global modification is realized at the end of each parallel cycle. The algorithm is carried out on DeepSuper-21C supercomputer, with 256 P4 Xeon 3.06/2.8 GHz Myrinet using MPI. Both the pixel signal-to-noise ratio (PSNR) for the decoded image and the speedup and efficiency for the parallel strategy are used for the evaluation of the proposed algorithm. Experimental results show that the performance of the algorithm improves by 0.1~0.2 dB with the execution time decreased considerably to 2-3 minutes.