基于改进残差神经网络和稀疏表示的脑部图像融合  

Brain image fusion based on improved residual neural network and sparse representation

作　　者：张亚加 黑荣婷 海梅 刘亚基 邵建龙[2] ZHANG Yajia;HEI Rongting;HAI Mei;LIU Yaji;SHAO Jianlong(College of Urban Construction,Yunnan Open University,Kunming 650500,China;Faculty of Information Engineering and Automation,Kunming University of Science and Technology,Kunming 650504,China)

机构地区：[1]云南开放大学城市建设学院,云南昆明650500 [2]昆明理工大学信息工程与自动化学院,云南昆明650504

出　　处：《陕西理工大学学报(自然科学版)》2025年第2期91-100,共10页Journal of Shaanxi University of Technology:Natural Science Edition

基　　金：国家自然科学基金项目(61302042);云南开放大学科学研究基金专项课题(23YNOU43)。

摘　　要：为了更好地提取融合图像的细节特征,保留更多有用信息,提高病灶结构检测能力,进而提高融合结果的清晰度,增强视觉效果,降低时间成本,提出了一种结合改进残差网络和稀疏表示的图像融合算法。使用边缘保持滤波器组分解源图像获得高、低频分量;对于稀疏性较差且含有较多结构信息的低频分量,设计了一种多范数加权度量的稀疏表示进行处理;对于含有较多纹理细节的高频分量,使用上下文转换模块对残差网络进行改进,提高特征提取的能力;最后,重构得到融合结果。从主观视觉和客观评价指标两个维度对结果进行综合评估,对比另外4种主流的融合方法,新推出的方法能够提高特征提取能力,保留更多有用的细节信息,增强了刻画病灶结构的能力,突出了病灶信息,在视觉效果和客观指标上均有显著优势,能够较好地为临床诊断、教学起到辅助作用。In order to better extract the detailed features of fused images,retain more useful information,improve the ability to detect lesion structures,thereby improving the clarity of fusion results,enhancing visual effects,and reducing time costs,the sudy has proposed an image fusion algorithm that combines improved residual network and sparse representation.Firstly,the study uses an edge preserving decomposition filter bank to decompose the source image and obtain high and low frequency components;For low-frequency components with poor sparsity and significant structural information,a multi norm weighted sparse representation for processing is designed;For high-frequency components with more texture details,a context conversion module is used to improve the residual network and enhance its feature extraction ability;Finally,the fusion result is obtained through reconstruction.The results are comprehensively evaluated from two dimensions:subjective visual and objective evaluation indicators.Compared with the other four mainstream fusion methods,the newly proposed method in this article can improve the ability of feature extraction,retain more useful detail information,enhance the ability to depict lesion structure,highlight lesion information,and have significant advantages in both visual effects and objective indicators.It can play a good auxiliary role in clinical diagnosis and teaching.

关 键 词：改进残差神经网络 稀疏表示 上下文转换模块 多范数加权度量 特征提取 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]