一种密度聚类的染色精子医学图像识别算法

摘要

当数据的密度有变化时，密度聚类算法 DBSCAN 不能一次发现多种密度的簇结构，通常需要调整参数，而合适参数的确定比较困难。提出了一种两阶段的密度聚类算法以识别精子图像，算法采用相同的参数完成对不同密度簇的发现。首先对原始数据图像采用初步的预处理技术，调用 DBSCAN 算法得到精子头部簇，然后对其余图像部分进行增强，以精子头部为核心点集合，再次调用 DBSCAN 算法得出密度可达的簇，从而完成精子图像的识别。实验证明对精子图像识别的准确率、效率、效果均优于传统密度聚类方法，为医生诊断病情提供有效的数据支持。%Density clustering algorithm DBSCAN is unable to find all clusters with various densities in a single run, so it is a trivial way to adjust parameters for various density clusters. However,suitable parameters are difficult to find. A new two-stage DBSCAN algorithm is described in processing sperm medical images,which can find clusters with various densities with the same parameter setting. Firstly,digital image preprocessing algorithms are used to help DBSCAN algorithm to find head of sperm clusters,and then the rest image is enhanced and density-reachable clusters are found by employing the head of sperm clusters as core clusters and running DBSCAN again to recognize the target sperm objects in the image. It is show that the accuracy rate,efficiency and effect of the present approach in image recognition is superior to those of the traditional one.