基于高斯混合模型的结核菌图像检测

Detection of Tuberculosis in Sputum Smear Images by Gaussian Mixture Models

结核病是严重危害人类健康的一类疾病。通过计算机图像处理手段进行自动检测结核菌计数可以大幅提高医生诊断效率。高斯混合模型是单一高斯分布的延伸,是使用多个高斯分布加权来拟合给定的数据样本,通过确定拟合参数确定每个样本的分类概率。该文首先通过向量量化算法对图像预处理,降低所需处理数据量,然后从HSV、CIEL＊a＊b＊、YCbCr颜色空间提取特征分量并送入高斯混合模型进行训练。根据实验结果,高斯混合模型比其他无监督分类算法（如K-means算法）准确度更高,与有监督的分类算法（如朴素贝叶斯分类算法）相比可以简化训练样本的制作,具有一定优势。

Cell recognition plays an important role in medical image-processing.First,we preprocess the images with vector quantization algorithm to reduce the computation.Then we extract different feature channels from HSV,CIEL＊a＊b＊ and YCbCr color spaces and put them into a Gaussian mixture model.Gaussian mixture models is a mature method for clustering unknown data.To determine the parameters of GMM,we use expectation maximization algorithm,which uses unlabeled data for model training.The experiment shows GMM finished the initial work of TB detection,while its performance wasn＇t high enough.