为了建立描述自然纹理的参数体系,选用木材这种典型自然纹理为研究对象.提取了木材纹理的5阶GMRF参数,为了降低运算量,采用改进的模拟退火算法进行参数的优化与选择,形成了描述木材纹理最优GMRF参数体系,并将其送入分类器进行分类识别.实验结果表明:集成神经网络的总体分类识别率为94.0％,近邻分类器的总体识别率为91.0％,获得了较高的分类识别率.说明用该参数体系对木材纹理进行分类识别是可行的,该参数体系也可用于与木材纹理相近的自然纹理的描述.%In order to establish the parameters system to describe natural texture,wood texture is adopted as the classic research objects.Parameters of 5-rank GMRF are extracted and parameters optimization and selection are carried on with an improved simulated annealing algorithm to reduce the computational complexity.So the parameter system based on GMRF is constructed and put into classifier to carry on classification.The result is that the classification and recognition ratio of integrated neural network reaches to 94.0% and the near neighbor classifier reaches to 91.0%,which indicate that realizing the classification and recognition of wood texture based on this GMRF parameter system is feasible.This parameter system can also be used to describe natural wood texture which is similar to wood texture.
展开▼
