结合线性稀疏表示和图像抠图的多聚焦图像融合方法  

Multi-focus image fusion method combining linear sparse representation and image matting

作　　者：王纪委 曲怀敬[1] 魏亚南 徐佳 张志升 张汉元 Wang Jiwei;Qu Huaijing;Wei Yanan;Xu Jia;Zhang Zhisheng;Zhang Hanyuan(School of Information&Electric Engineering,Shandong Jianzhu University,Jinan 250101,China)

机构地区：[1]山东建筑大学信息与电气工程学院,济南250101

出　　处：《计算机应用研究》2022年第6期1879-1885,1890,共8页Application Research of Computers

基　　金：国家自然科学基金资助项目(62003191);山东省自然科学基金资助项目(ZR2014FM016)。

摘　　要：针对现有的多聚焦图像融合方法对聚焦/散焦边界(FDB)信息捕捉不准确的问题,提出了一种新的基于线性稀疏表示和图像抠图的多聚焦图像融合方法。首先,引入一种基于线性稀疏表示的焦点测度,它利用自然图像形成的字典与输入图像在局部窗口上的线性关系,通过求解线性系数来表示图像的焦点信息。然后,利用焦点测度获取源图像的焦点图和一个由聚焦区域、散焦区域以及包含FDB的未知区域组成的三元图,并将三元图作为一个输入,采用图像抠图技术处理源图像的FDB区域,从而得到较精确的全聚焦图像。最后,为了进一步提高融合图像的质量,将得到的全聚焦图像作为新字典实现融合过程的迭代进行,在经过设定的更新次数后得到最终的全聚焦融合图像。实验结果表明,相比于11种最先进的多聚焦图像融合方法,该方法具有较好的融合性能和视觉效果,且有较高的计算效率。Aiming at the problem of inaccurate information capture of focused/defocused boundary(FDB)by existing multi-focus image fusion methods,this paper proposed a new multi-focus image fusion method based on linear sparse representation and image matting.Firstly,the paper introduced a focus measure based on linear sparse representation,which used the linear relationship between the dictionary formed by the natural image and the input image in the local window to represent the focus information of the image by solving the linear coefficient.Then,it used the focus measure to obtain the focus map of the source image and a ternary map consisting of the focus area,the defocus area,and the unknown area including the FDB,further used the image matting technology to process the ternary image,to obtain a more accurate full-focus image.Finally,to further improve the quality of the fused image,it used the obtained full-focus image as a new dictionary to realize the iterative process of fusion,and obtained the final full-focus fused image after the set update times.Experimental results show that compared with the 11 most advanced multi-focus image fusion methods,the proposed method has better fusion performance and visual effect,and higher computational efficiency.

关 键 词：多聚焦图像 图像融合 焦点测度 线性稀疏表示 图像抠图 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]