一种基于变分方法的多分辨率域融合策略  

Variational Fusion Strategy for Multiscale Analysis Domain

作　　者：马宁[1,2] 周则明[1] 罗立民[2] 

机构地区：[1]解放军理工大学气象海洋学院,南京210007 [2]东南大学影像科学与技术实验室,南京210096

出　　处：《计算机科学》2013年第9期270-274,共5页Computer Science

基　　金：国家自然科学基金(41174164);解放军理工大学预研基金(KYQYZLXY1207)资助

摘　　要：针对常规多分辨率域融合策略的处理过于直接可能会降低融合图像质量的现象,提出了一种多分辨率域变分融合策略。该策略根据源图像分解系数的梯度信息构造目标系数梯度场,寻找梯度最接近目标梯度场的系数作为融合图像的分解系数。为了保持与源图像的相似性,对高、低频系数分别加入了约束条件。针对高频系数对噪声敏感的特点,还引入了全变分项来保证解的平滑性。在多聚焦图像、红外图像与可见光图像上的融合实验结果表明:与常规融合策略相比,提出的融合策略能够注入更多的空间细节信息,同时能更有效地保留源图像的结构信息。Conventional fusion strategies typically manipulate the fused coefficiented too directly, which may introduce artifacts and decrease the fused image quality. This paper presented a variational fusion strategy for multiscale analysis domain. An energy minimization problem was proposed to find the solution whose gradient is closest to that of the tar-get gradient field, which is built from the gradient information of the source images＇ decomposition coefficients. To keep the similarity with source images, the constraint condition was imposed into the highpass and lowpass coefficients sepa- rately. Since the highpass coefficients are sensitive to the noise, the total variation term is also introduced to assure the smoothness of the solutiorL The performance of the proposed fusion strategy was evaluated on multifocus images, infra- red and visible images. Experimental results show that the strategy can inject more spatial detail information and mean- while preserve more structural information of the original image than conventional fusion strategies.

关 键 词：多分辨率分析 变分 图像融合 融合策略 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]