骨髓细胞图像的小波变换与K-means聚类分割算法-《郑州大学学报(工学版)》

文章信息/Info

Title:: Segmentation Algorithm for Bone Marrow Cell Image Based on TheWavelet Transform and K-means Clustering

Author(s):: SU Shi mei1; LV Xue yang2; School of Electrical Engineering ,Zhengzhou University ,Zhengzhou 450001，China

Keywords:: cell segmentation; wavelet transform; K-means clustering; color feature

摘要:: 为了能准确地分割出骨髓细胞涂片中的各类细胞,提出一种基于小波分析的聚类分割方法．首先采用小波变换消除散焦噪声,然后通过对彩色图像G分量进行小波系数多尺度分解,提取特征参数信息,根据图像G分量与S分量的差异性并结合得到变换图像STG,二值化处理提取白细胞胞核,最后为K-means聚类方法提供优化的初始聚类中心,从而对各类红细胞、白细胞进行分割和分离.通过对比分析和实验测试,该算法有效克服了骨髓细胞显微图像的复杂散焦、细胞种类繁多以及目标区分度低而导致图像分割的困难,准确率达94.15% .

Abstract:: A segmentation method based on the wavelet analysis is proposed in this paper for segmenting visi-ble components from the bone marrow cell. Firstly,the wavelet transform is used to erase the effect of defocus-ing.Secondly,G component is chosen as clustering the information of characteristic parameters by using thethreshold multi-scale wavelet analysis,which is combined with saturation component according to different dis-tribution characteristic of leukocyte nucleus to construct a transpositional image named STG. The nucleus areextract by thresholding the image.Ultimately,it can provide an optimal initial clustering centers for K-meansclustering method to segment the red blood cells and the white blood cells from the background as well as sepa-rate them from each other. Through comparative analysis and algorithm testing,this method can effectively o-vercome the difficulties of complex components,poor discrimination and complicated defocusing during thecells segmentation ,the accuracy rate reaches 94.15 % .