利用一致性低秩表示模型进行多聚焦图像融合

Multi-focus Image Fusion Using Consistency Low Rank Representation Model

多聚焦图像融合是图像融合的一个重要分支,它能够将多幅源图像中的清晰区域提取出来,融合成一幅清晰度较高、质量较好的图像。提出一种新的多聚焦图像融合方法,将低秩模型和空间邻域一致性有效地结合起来,克服现有一些方法所存在的问题。主要采用低秩表示模型,将源图像分解为背景信息和细节信息,并用结构丰富的细节信息作为聚焦性测量进行多聚焦图像融合。其次,为了提高融合质量,缓解“块”状模糊,采取了空间一致性约束,将拉普拉斯矩阵作为约束项构建出新的低秩表示模型并进行了求解。实验表明,该多聚焦融合算法的融合性能优于现有的其他一些算法。

Multi-focus image fusion is an important branch of image fusion. It can extract clear regions from multiple source images and fuse them into an image with higher clarity and better quality. This paper presents a novel multi-focus image fusion method, efficiently combining low-rank representation (LRR) and spatial consistency to avoid the problems in some existing methods. In the paper, the low rank representation model is used to decompose the source images into the background information and the detail information. Then the image fusion is addressed by employing the detail information in which with the rich structures as the focusing measurement. Secondly, in order to improve the quality of fusion and alleviate the blocking-artifacts, the spatial consistency constrained is adopted in this paper. The Laplacian matrix is used as a constraint item to construct a new low rank representation model and to have been solved. Experimental results demonstrate that the multi-focus image fusion algorithm proposed in this paper consistently outperforms existing state-of-art methods in terms of both visual and quantitative evaluations.