白带常规中显微图像细胞的自动识别一直是悬而未决的难题。上皮细胞是白带显微图像中的主要有型成分, 能够直接反应清洁度等指标。针对目前白带常规中人为主观判断效率低的特点, 本文提出了一种基于纹理特征的白带显微图像中上皮细胞检测方法。首先, 应用形态学方法实现对上皮细胞等前景目标的提取; 其次, 分析前景目标的局部二值模式纹理特征; 最后, 用支持向量机实现对上皮细胞的精确分类。实验证明, LBP纹理特征在上皮细胞的检测和识别方面较其他的纹理特征提取器均取得了很好的检测效果, 精确率为89.5%, 召回率为86.0%。检测效率高, 检测时间为304 ms。本文算法已经应用于临床测试中, 并取得了很好的临床实验效果。

The automatic recognition of microscopic image cells in leucorrhea routine has been an unsolved problem. Epithelial cells are the main visible components in leucorrhea microscopic images, which can reflect the indicators such as cleanliness directly. Aiming at the low efficiency of subjective judgment in leucorrhea routine, a new method for detecting epithelial cells in leucorrhea microscopic images based on texture features was proposed. Firstly, the morphological method was used to extract the foreground target such as epithelial cells. Then, the texture features of the foreground target were analyzed based on local binary pattern. Finally, the support vector machine was used to classify epithelial cells accurately. The experimental results show that LBP achieves better detection results than other texture feature extractors in recognition precision and recall rate with the precision of 89.5% and recall of 86.0%. The detection efficiency is high and the detection time is 304 ms. The proposed algorithm has been applied to clinical testing, and achieved good clinical experimental results.