血液白细胞计算机分类中的特征提取研究

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。

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

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.