ECONOMIZING ENHANCED FUZZY MORPHOLOGICAL ASSOCIATIVE MEMORY
Enhanced Fuzzy Morphological Associative Memory (EFMAM) successfully conquers the common obstacle of MAM and FMAM, i.e. the extreme vulnerability to the hybrid noise. However, as the number of training patterns increases, EFMAM encounters difficulties in hardware realization, because its network architecture becomes larger and larger. Meanwhile its space and time complexity also rapidly increase. The reason consists in the un-economization of Empirical Kernel Map (EKM) vectors in EFMAM. In this paper, we propose an economized EFMAM, called E2FMAM, which first define a criterion to economize EKM vectors, then use the famous Genetic Algorithms (GAs) to search the optimum. The simulation results show that E2FMAM has less space and time complexity than EFMAM, and a comparable recognition performance to EFMAM in terms of the tolerance to different types and levels of noise or information incompletion. Besides, its inscnsitivity to image resolution brings us the flexibility in the higher-resolution image recognition problem.
Associative memory Empirical kernel map Fuzzy Morphological neural networks
MIN WANG RONG CHU
College of Computer Science & Engineering, Hohai University, Nanjing 210098, China College of Electrical Engineering, Hohai University, Nanjing 210098, China
国际会议
2008 International Conference on Machine Learning and Cybernetics(2008机器学习与控制论国际会议)
昆明
英文
495-500
2008-07-12（万方平台首次上网日期，不代表论文的发表时间）