패턴분류기의 설계 및 분석

Abstract

The important prerequisites of designing the pattern classifier in formation age are high data throughput and small storage space and low cost hardware implementation. In this thesis, we propose a method for designing LFSR(Linear Feedback Shift Register) based pattern classifier, and we propose a method for designing TPMACA(Two Predecessor Multiple Attractor Cellular Automata) based pattern classifier. The propose pattern classifier based on LFSR is easy to see non-reachable state and we can obtain (Dependency Vector) by using the -basic path. Also, we propose a method for generating pattern classifier based on LFSR which correspond to given dependency vector. The MACA(Multiple Attractor Cellular Automata) is adapted for use as a pattern classifier. We propose a method of finding for classifying the patterns in a database. And we propose various methods for the construction of the TPMACA with two attractor trees by using . Non-linear CA is difficult to analyze than the linear CA. So it is difficult to identify reachable states and attractors of nongroup CA than linear CA. So in this thesis, we propose reachable table to overcome these problems. We can see the next state for all the states of the non-linear CA by the proposed reachable table. Also, we can identify next state for all the states of the non-linear CA. In addition, we can identify reachable states and attractors by reachable table.