计算机应用 ›› 2016, Vol. 36 ›› Issue (11): 3188-3195.DOI: 10.11772/j.issn.1001-9081.2016.11.3188

• 虚拟现实与数字媒体 • 上一篇    下一篇

基于Curvelet变换和多目标粒子群的混合熵MRI图像多阈值分割

卞乐1,2, 霍冠英1,2, 李庆武1,2   

  1. 1. 河海大学 物联网工程学院, 江苏 常州 213022;
    2. 常州市传感网与环境感知重点实验室, 江苏 常州 213022
  • 收稿日期:2016-05-09 修回日期:2016-06-14 出版日期:2016-11-10 发布日期:2016-11-12
  • 通讯作者: 霍冠英
  • 作者简介:卞乐(1991-),女,江苏盐城人,硕士研究生,主要研究方向:数字图像处理;霍冠英(1979-),男,河南汝南人,副教授,博士,主要研究方向:声呐图像处理;李庆武(1964-),男,河南新乡人,教授,博士生导师,博士,主要研究方向:数字图像处理。
  • 基金资助:
    国家自然科学基金资助项目(41306089);江苏省自然科学基金资助项目(BK20130240)。

Multi-threshold MRI image segmentation algorithm based on Curevelet transformation and multi-objective particle swarm optimization

BIAN Le1,2, HUO Guanying1,2, LI Qingwu1,2   

  1. 1. College of Internet of Things Engineering, Hohai University, Changzhou Jiangsu 213022, China;
    2. Changzhou Key Laboratory of Sensor Networks and Environment Sensing, Changzhou Jiangsu 213022, China
  • Received:2016-05-09 Revised:2016-06-14 Online:2016-11-10 Published:2016-11-12
  • Supported by:
    This work is partially supported by the National Natural Science Foundation of China (41306089), the Natural Science Foundation of Jiangsu Province (BK20130240).

摘要: 针对因噪声干扰多、灰度不均匀、目标边界模糊导致的核磁共振成像(MRI)图像难以精确分割的问题,提出了一种基于Curvelet变换和多目标粒子群(MOPSO)的混合熵MRI图像多阈值分割算法。首先,对待分割MRI图像进行Curvelet分解,提取低频子带和高频细节子带构建概貌-细节灰度级矩阵模型,以提高算法的目标细节表示能力;其次,同时考虑目标与背景的类间差异性与类内均匀性,将提出的二维多阈值倒数熵和倒数灰度熵组合定义为混合熵,作为多目标粒子群算法的目标函数,协同搜索得到最优的分割多阈值,以实现MRI图像的精确分割;最后,为提高算法的求解速度,提出了二维倒数熵和倒数灰度熵多阈值选取的梯度算法。实验结果表明:与二维tsallis熵、自动细菌觅食分割法(ABF)和改进的Otsu多阈值分割算法相比,所提方法对灰度不均和含噪的MRI图像具有更好的适应性,分割结果更为精确。

关键词: 核磁共振成像, Curvelet变换, 多目标粒子群, 二维倒数熵, 二维倒数灰度熵

Abstract: To deal with the difficulties caused by noise disturbance, intensity inhomogeneity and edge blurring in Magnetic Resonance Imaging (MRI) image segmentation, a new multi-threshold MRI image segmentation algorithm based on mixed entropy using Curvelet transformation and Multi-Objective Particle Swarm Optimization (MOPSO) was proposed. First, the high-frequency and the low-frequency subbands were obtained using Curvelet decomposition, which were used to construct the profile-detail gray level matrix model that could represent edge details accurately. Then, with the consideration of both inter-class similarity and intra-class difference of background and object region, two-dimensional reciprocal entropy and reciprocal gray entropy were proposed and combined to define the mixed entropy, which was used as the objective function of MOPSO. The optimal multi-threshold was searched cooperatively to get an accurate segmentation. Finally, in order to speed up the segmentation process, gradient-based multi-threshold estimation algorithms for two-dimensional reciprocal entropy and reciprocal gray entropy were proposed. The experimental results show that the proposed method is more adaptive and accurate when applied to gray uneven and noisy MRI image segmentation in comparison with two-dimensional tsallis entropy, Adaptive Bacterial Foraging (ABF) and improved Otsu multi-threshold segmentation algorithms.

Key words: Magnetic Resonance Imaging (MRI), Curvelet transformation, Multi-Objective Particle Swarm Optimization (MOPSO), two-dimensional reciprocal entropy, two-dimensional reciprocal gray entropy

中图分类号: