基于优选特征属性偏序结构分析的白细胞图像分类规则发现

Classification rule discovering for leucocyte images based on analysis of preselected features' attribute partial-ordered structure

基于形式概念分析和属性偏序结构理论,提出了一种白细胞图像分类规则发现模式,从而建立了高效的白细胞图像分类方法。用该方法,首先在大量的白细胞图像区域特征测定实验基础上对白细胞图像优选特征进行了离散化处理,针对白细胞图像数据集构建了形式背景,依据分层类坐标矩阵的属性偏序结构生成方法生成了白细胞图像数据集属性偏序结构图;然后通过对属性偏序结构图分析,发现了6类白细胞相应的6条分类规则;最后,依据分类规则建立了二分树分类器,并且在实际白细胞图像数据集测试实验中取得了94.04%的平均分类精度,该精度明显高于其它3种经典算法,证明了基于优选特征属性偏序结构分析获取的白细胞图像分类规则的可用性、简单性和有效性。

A novel model for finding leucocyte image classification rules was proposed based on formal concept analysis and the theory of attribute partial-ordered structure, and then, an efficient method for leucocyte image classification was established. According to the method, the optimized leucocyte attributes were discretized based on the analysis of the obtained experimental measurement results, and the optimized formal concept and the attribute partial-or- dered structure were established based on the hierarchical class coordinate matrix for actual leucocyte images data- set. Then, based on the optimized attribute partial-ordered structure, six classification rules for six kinds of leuko- cytes were extracted. Finally, a binary decision tree classifier was established according to the classification rules, and an average classification accuracy of 94.04% was achieved. The classification accuracy is significantly higher than the other 3 kinds of classical algorithms, showing the better usability, simplicity and effectiveness of the classi- fication rules obtained based on analysis of the attribute partial-ordered structure.