八通道MSVD构造及其在多聚焦图像融合中的应用  

Construction of Eight Channel Multi-Resolution Singular Value Decomposition of Matrix and Its Application in Multi-Focus Image Fusion

作　　者：刘斌[1] 刘维杰[2] 罗益辉[1] 郭琳[1] 

机构地区：[1]湖北大学计算机与信息工程学院,湖北武汉430062 [2]武汉大学计算机学院,湖北武汉430072

出　　处：《电子学报》2016年第7期1694-1701,共8页Acta Electronica Sinica

基　　金：国家自然科学基金面上项目(No.61471160);湖北省自然科学基金重点项目(No.2012FFA053)

摘　　要：针对经典的SVD在图像处理中的不足,提出了一种八通道多尺度奇异值分解(Multi-resolution Singular Value Decomposition,MSVD)构造方法,并把它应用于多聚焦图像融合中.首先,在经典SVD的基础上,利用矩阵分块的方法,提出了一种八通道多尺度SVD的构造方法.其次,对参加融合的多聚焦图像进行八通道MSVD分解,得到高层低频和各层七个方向的高频,对分解的低频子图像利用数学形态学增强边缘的方法进行融合、高频子图像采用基于区域能量取大的融合规则进行融合,并重构获得融合结果图像.最后,对融合结果进行主客观评价和分析.实验结果表明,该图像融合方法有较好的视觉效果,结果图像有较高的清晰度,边缘细节信息丰富,没有方块效应.从客观数值和图形评价指标看,该方法有较高的清晰度,其清晰度比基于DWT的融合方法、基于LWT的融合方法、基于Curvelet的融合方法、基于Contourlet的融合方法都高.To improve the defaults of classical SVD in image processing,a construction method of eight channel multi-resolution singular value decomposition of matrix （MSVD）is presented.An image fusion method based on this MS-VD is proposed.Firstly,based on the principle of classical SVD and blocking algorithm,a multi-resolution analysis of eight-channel SVD of matrix is constructed.Each image involved in the fusion are decomposed into one approximation and seven detail images with different resolution by the eight channel multi-resolution singular value decomposition.Secondly,com-bined with reconstruction algorithm of MSVD,the frame of image fusion is given.The different frequency of original images can be shown in multi-resolution form.The low-frequency sub-image is fused by using an edge enhancement method of mathematical morphological gradient.For the seven high-frequency sub-images of each level,the energy of each image patch over 3 ×3 window in the high-frequency sub-images is computed as activity measurement.The center pixel of the 3 ×3 win-dow in which the energy is bigger is selected as the new pixel of the fused result images.Finally,the performance of the re-sult image is evaluated using objective numerical and graphics indices.The experimental results show that the proposed method has good visual effect and has no blocking-artifact.When compared with the fusion method based on DWT,LWT, Curvelet and Contourlet,the proposed fusion method has been observed to have higher definition.

关 键 词：图像融合 矩阵奇异值分解 多尺度分析 多聚焦图像 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]