基于复剪切波变换与VGG19模型的医学图像融合方法  

The medical image fusion method based on the complex shearlet transform and the VGG19 model

作　　者：王钰帏 王雷[1] 郭新萍 程天琪 WANG Yuwei;WANG Lei;GUO Xinping;CHENG Tianqi(School of Computer Science and Technology,Shandong University of Technology,Zibo 255049,China)

机构地区：[1]山东理工大学计算机科学与技术学院,山东淄博255049

出　　处：《山东理工大学学报（自然科学版）》2024年第4期53-60,共8页Journal of Shandong University of Technology:Natural Science Edition

基　　金：山东省自然科学基金项目(ZR2021MF017)。

摘　　要：针对传统医学图像融合方法存在的细节信息不够清晰、边缘信息易丢失和图像失真等缺点,以及深度学习网络缺乏足够的训练数据集等问题,提出了一种基于复剪切波变换和预训练网络模型VGG19的多模态医学图像融合方法。首先,利用复剪切波变换提取医学图像边缘和纹理信息,并得到多尺度、多方向的子带系数。然后,使用加权局部能量和修正的拉普拉斯算子对低频子带系数进行融合;引入预训练的VGG19提取多层特征图,结合加权评估规则来获取高频子带的融合结果。最后,对融合的高频和低频子带,施加复剪切波逆变换重构融合图像。实验表明,该方法得到的融合图像,不仅可以清晰地显示图像的细节信息和边缘信息,而且能够有效抑制伪影和失真现象的产生,在主观视觉比较和6种客观评价指标下能够达到更佳融合效果。To deal with the drawbacks of traditional medical image fusion methods,such as the insufficient clarity of fine details,the loss of edge information,and the image distortion,as well as the insufficient training datasets for deep learning,a multi-modal medical image fusion method based on the complex shearlet transform(CST)and the pre-trained VGG19 network model is proposed.The CST is firstly employed to extract the edge and texture information,which are represented by the multi-scale and multi-directional sub-band coefficients.Then,the low-frequency sub-band coefficients are fused using the weighted local energy and a modified Laplacian operator.The pre-trained VGG19 model is introduced to extract the multi-layer feature maps,which are combined with the weighted evaluation rules to obtain the fused the high-frequency sub-bands.Finally,the fused high-frequency and low-frequency sub-bands are treated by the inverse transform of the CST to reconstruct the fused image.Experimental results demonstrate that the proposed method produces the fused images that not only display clear details and contours but also effectively suppress the generation of artifacts and distortion.Better fusion performance can be achieved by the proposed method via subjective visual comparison and six objective evaluation metrics.

关 键 词：医学图像 图像融合 复剪切波变换 VGG19模型 修正的拉普拉斯算子 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] R445[自动化与计算机技术—计算机科学与技术]