【摘要】 结核杆菌的有效识别是实现结核杆菌显微图像自动分析的关键。论文针对萋尔-尼尔逊方法染色的显微图像,对结核杆菌的提取和识别进行研究。算法首先基于形态学top-hat变换对图像进行增强并进行阈值分割,然后对结核杆菌进行特征分析,定义并提取了包括颜色、形状和边界不规则性等在内的10个特征,最后运用BP神经网络进行识别。实验结果表明,该方法能够有效检测并识别结核杆菌。选取1 200个目标进行七折交叉验证,识别准确率达到98%。更多还原

【Abstract】 Effective identification of mycobacterium tuberculosis is the key to automatic analysis of mycobacterium tuberculosis in microscopic image.The algorithm of the extraction and identification of mycobacterium tuberculosis in the microscopic images of the Ziehl-Nelson stained bacteria were investigated in this study.Firstly,the microscopic images were enhanced using the morphology top-hat transform and segmented based on a threshold,and then the features,including color,shape and border irregularity were defined and calculated.Finally the mycobacterium tuberculosis was identified by BP neural network.The experiment results show that mycobacterium tuberculosis can be effectively detected and identified using the algorithm proposed in this paper.The accuracy has reached 98% using sever fold cross-validation on 1200 selected targets.更多还原