血液白细胞计算机分类中的特征提取研究

STUDY OF LEUKOCYTE FEATURES IN PERIPHERAL BLOOD FOR AUTOMATIC CLASSIFICATION YANG XIAOMIN

在所建的血液白细胞自动分类ＬＥＵＫ分析系统上，对每个细胞提取了不同性质的三组共３５个特征，包括７个形状参数、１２个光密度参数和１６个纹理参数。对１２００余六类白细胞样品进行了自动分类测试。实验结果表明：若仅用形态和光密度特征，总体识别率达８０％左右，其主要误识率来自于颗粒细胞的分类。而该文提出的纹理特征弥补这一不足，使系统的总体识别率提高了十个百分点。该文重点介绍这三组特征，并就其在白细胞自动识别分类中的作用进行详细讨论。

For each cell,a total of 35 features in three kinds are extraeted from theautomatic classification system of peripheral blood leukocytes.They are:7 morpho-logic features,12 optical densiey features and 16 texture features.The testing for1200 leukocytes indicates that using these three kinds of features is better thanjust using the first 2 kinds,and the accuracy rate is approximately 90%. The resultis discussed.