【摘要】 应用统计模式识别的方法对细胞图象进行计算机自动识别时，细胞特征的抽提与选用是重要的一环。我们在所建的细胞图象自动分析系统上对每个细胞抽提了不同性质的三组特征共 ３３个。即 ２０个综合性特征、 １０个统计性直方图特征和 ３个颜色特征。并对约２００个呼吸道（肺）细胞中的重度不典型化生细胞和高分化鳞癌细胞进行自动分类。试验结果证明：综合选用三组特征比仅选用其中的一组或两组特征的效果好。正确识别率达８０％左右。本文重点就三组特征的性质及其在细胞自动识别与分类中的作用作一初步的分析。更多还原

【Abstract】 Automatic recognition and classification utilizing statistical pattern recognition method for cell images can be performed by computers, and the feature extraction and selection are considered to be the key points of this method. For each cell, three different feature sets have been extracted by a cell image analysis system. Three feature sets were adopted, including 20 global features, 10 statistical histogram features and 3 color features. About 200 lung cells have been classified into two classes: marked atypical squamous metaplasia cells and well-differention squamous carcinoma cells. As a result of the classification, it has been proved that the result obtained by the utilization of all the three feature sets is rather better than that as the utilization of one or two of them, the number of cells which agree with the visual assignment can be obtained more than 80%. In this paper, the nature of the three feature sets and their effects on the classification results have been mainly analyzed.更多还原