中国组织工程研究

中国组织工程研究 ›› 2010, Vol. 14 ›› Issue (13): 2365-2368.doi: 10.3969/j.issn.1673-8225.2010.13.022

• 骨与关节图像与影像 bone and joint imaging • 上一篇    下一篇

基于多重分形谱和自组织神经网络的医学图像分割

金春兰，黄  华，张国芳，刘圹彬   

  1. 四川大学电气信息学院，四川省成都市  610065
  • 出版日期:2010-03-26 发布日期:2010-03-26
  • 通讯作者: 黄 华，博士（后），教授，博士生导师，四川大学医学信息工程系，四川省成都市 610065 hhua@scuedu.cn
  • 作者简介:金春兰★，女，1985年生，朝鲜族，吉林省延吉市人，四川大学在读硕士，主要从事医学图像处理与医学信号处理研究。 Xiaoai20008@163.com

Medical image segmentation based on multi-fractal spectrum and self-organizing neural network

Jin Chun-lan, Huang Hua, Zhang Guo-fang, Liu Kuang-bin   

  1. School of Electrical Engineering and Information, Sichuan University, Chengdu  610065, Sichuan Province, China
  • Online:2010-03-26 Published:2010-03-26
  • Contact: Huang Hua, Doctor, Professor, Doctoral supervisor, School of Electrical Engineering and Information, Sichuan University, Chengdu 610065, Sichuan Province, China hhua@scuedu.cn
  • About author:Jin Chun-lan★, Studying for master’s degree, School of Electrical Engineering and Information, Sichuan University, Chengdu 610065, Sichuan Province, China Xiaoai20008@163.com

PDF

1885

被引次数

20

摘要：

背景：单一的多重分形谱图像分割虽然在边缘及纹理的区分上有较大优势，但是选择不同的测度，不同的阈值对于分割结果影响比较大，正确地选择最优的测度比较困难。
目的：结合多重分形谱图像分割法及自组织特征映射神经网络对医学图像进行处理。
方法：以图像每一象素及其周围象素的均值及方差为基本特征，再结合4种不同多重分形谱为纹理特征，实现自组织特征映射神经网络。
结果与结论：选择不同的测度对同一图像的分割结果是不一样的，并且同一种测度对不同图像的分割效果也不一样，说明基于多重分形谱的医学图像分割中选择合适的测度是一个关键所在。因此将多重分形谱结合自组织特征映射神经网络的方法对图像进行处理，该方法省略了选择测度的步骤，直接把4种多重分形谱作为特征，与另两种基本特征一起作为自组织神经网络的输入，对网络进行学习，并自动对图像进行分割。实验结果表明该方法在满足复杂图像中有效进行分割的同时达到了自动、自适应的目的。

关键词: 多重分形, 自组织特征映射神经网络, 医学图像分割, 纹理, 数字化影像技术

Abstract:

BACKGROUND: Though the sole multi-fractal spectrum image segmentation has a great advantage when distinguishing edge texture, the results are influenced to be different when adopting different measures and thresholds when adopting different measures. So it is difficult to correctly adopt optimal measure.
OBJECTIVE: Combing the multi-fractal spectrum image segmentation method and self-organizing feature map neural network to process the image.
METHODS: The mean and variance of every pixel and others around were served as the basic characteristics. And the texture characteristics were combined with four different multi-fractal spectrums to realize self-organizing neural network.
RESULTS AND CONCLUSION: The results were different when adopting different measures and thresholds, meantime, the results were different to the different image using the same measure. So the key was how to choose the suitable measure. The method of multi-fractal spectrum image segmentation method binding self-organizing feature map neural network was used. It omitted the steps of choosing measures, and immediately made four multi-fractal spectrums be characteristics combining the other two characteristics to be the input of self-organizing neural network. Then the net was studied, and the image was automatically segmented. The experiment results showed that this method can fulfill effective segmentation in the complicated image with automatic and adaptively.

中图分类号: