基于图像分解和稀疏表示的多聚焦图像融合  

Multi-focus image fusion based on image decomposition and sparse representation

作　　者：蒋留兵 张点 车俐 Jiang Liubing;Zhang Dian;Che Li(School of Computer Science&Information Security,Guilin University of Electronic Technology,Guilin Guangxi 541004,China)

机构地区：[1]桂林电子科技大学计算机与信息安全学院,广西桂林541004

出　　处：《计算机应用研究》2021年第8期2499-2504,共6页Application Research of Computers

基　　金：国家自然科学基金资助项目(61561010);广西自然科学基金资助项目(2017GXNSFAA198089);广西重点研发计划资助项目(桂科AB18126003,AB18221016)。

摘　　要：针对多聚焦图像融合中缺乏细节保护和结构不连续的不足,提出了一种基于图像分解的多聚焦图像融合算法。首先,源图像采用卡通纹理图像分解得到卡通部分和纹理部分;其次,卡通部分采用卷积稀疏表示的方法进行融合,纹理部分采用字典学习进行融合;最后,将卡通和纹理部分融合得到融合图像。实验建立在标准的融合数据集中,并与传统和最近的融合方法进行比较。实验结果证明,该算法所获得的融合结果在方差和信息熵上具有更好的表现,该算法能够有效克服多聚焦图像融合中缺乏细节保护和结构不连续的缺点,同时有更好的视觉效果。Aiming at the lack of detail protection and discontinuous structure in multi-focus image fusion,this paper proposed a multi focus image fusion algorithm based on image decomposition and sparse representation.Firstly,this algorithm used cartoon and texture image decomposition to decompose the source images into the cartoon parts and the texture parts.Secondly,it used convolution sparse representation to fuse the cartoon parts,and used dictionary learning to fuse the texture parts.Finally,it fused the cartoon part and the texture part to obtain the fused image.The experimental results show that the proposed algorithm has better performance than other fusion method.It can effectively overcome the lack of detail protection and structural discontinuity in multi-focus image fusion and has better visual effect.

关 键 词：多聚焦图像融合 卡通纹理图像分解 卷积稀疏表示 字典学习 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]