一种适用于多模态医学图像融合的自适应脉冲耦合神经网络改进算法  

An Adaptive PCNN Improved Algorithm Suitable for Multi-modality Medical Image Fusion

作　　者：于淼 宁春玉[1] 石乐民 吕冰垚 YU Miao;NING Chun-yu;SHI Le-min;LV Bing-yao(School of Life Science and Technology, Changchun University of Science and Technology, Changchun 130022, China)

机构地区：[1]长春理工大学生命科学技术学院,长春130022

出　　处：《科学技术与工程》2020年第22期9116-9121,共6页Science Technology and Engineering

基　　金：吉林省科技发展计划(20180623008TC)。

摘　　要：针对医学图像融合存在伪影、边缘保持性弱等问题,提出了一种参数自适应的脉冲耦合神经网络(pulse coupled neural network,PCNN)图像融合方法。首先,对源图像通过非下采样Contourlet变换(non-subsampled contourlet transform,NSCT)得到一个低通子带和多个尺度多个方向下的带通子带。然后用区域标准差调整连接范围,进而调整突触权重矩阵以及加权系数;用各子带的改进空间频率中方向特征最显著的分量调整连接强度;对于外部激励,低通子带用区域能量和区域方差的线性组合计算,带通方向子带采用改进的拉普拉斯能量和计算。点火映射图的判决遵循取大原则。最后,通过NSCT逆变换得到融合结果图。实验结果表明,此算法能更多地保留源图像的信息,边缘保持能力更强,融合图像对比度高,视觉效果更佳,适用于多种模态医学图像之间的融合。Aiming at the problems of artifacts and weak edge preservation in medical image fusion,an image fusion algorithm based on parametric adaptive pulse coupled neural network(PCNN)was proposed.Firstly,the source image was transformed by non-subsampled Contourlet transform(NSCT)to get a lowpass subband and a bandpass subband with multiple scales and directions.Secondly,the connection range was adjusted by regional standard deviation,and then the synaptic weight matrix and weighting coefficient were adjusted.The connection strength was adjusted by improving the component with the most significant directional feature in the spatial frequency of each subband.For external excitation,the lowpass subband was calculated by the linear combination of regional energy and regional variance,and the bandpass directional subband was calculated by sum-modified Laplacian.The judgment of fire mapping image followed the maximum principle.Finally,the fusion image was obtained by NSCT inverse transformation.The experimental results show that the algorithm can retain more information of the source images with stronger edge retention ability.The fusion image has higher contrast and better visual effect.It is suitable for the fusion of many kinds of modal medical image.

关 键 词：图像处理 医学图像融合 自适应脉冲耦合神经网络 改进空间频率 区域特征 改进拉普拉斯能量和 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]