中文摘要：

针对传统纹理图像分类算法识别率不高的问题,引入置信规则库推理方法而提出一种纹理图像分类策略.目前纹理图像分类研究常局限于纹理特征提取算法的改进,而忽视了另一个决定分类效果的关键,即分类器设计.该文采用置信规则库推理方法,在现有纹理特征提取算法基础上重新设计纹理图像分类器.根据角度径向变换和灰度共生矩阵算法提取图像纹理特征,采用主成分分析方法减少角度径向变换特征的维数,以避免产生置信规则库“组合爆炸”的问题.最后用置信规则库推理方法将纹理特征信息转换成类别置信度信息,得到最终的分类结果.实验中将置信规则库推理方法分别与相似性距离度量法和支持向量机法进行对比,结果表明所提出的方法在一定程度上提高了纹理图像分类准确率.

英文摘要：

To improve precision of traditional texture image classify algorithm, a new texture image classification method based on belief rule-base inference methodology using evidential reasoning approach（RIMER） is proposed. Researches on texture image classifi- cation generally consider improving texture feature extraction, and the design of classifier that is crucial to classification precision is largely ignored. In this paper, a rule-base inference method using an evidential reasoning approach is proposed. The classifier is redesigned based on the current methods of texture feature extraction. Algorithms of angular-radialtransform and gray-level con-occurrence matrix are used to extract texture image feature. Principle component analysis is carried out to solve the problem that the size of a belief rule base（BRB） classifier is controlled within a feasible range. The approach of rule-base inference method with evidential reasoning transforms the texture features into classified belief degree information. Practicability and effectiveness of the proposed approach is validated in a case study.