基于视觉结构相似度的离焦图像融合方法  

A New Fusion Method for Lost-focus Images Based on Vision Structural Similarity

作　　者：曹寅[1] 蔡云泽[1] 韩瑜[1] 

机构地区：[1]上海交通大学电子信息与电气工程学院,上海200240

出　　处：《上海交通大学学报》2010年第11期1501-1506,共6页Journal of Shanghai Jiaotong University

基　　金：国家自然科学基金资助项目(61004088);上海市基础研究重点资助项目(09JC1408000)

摘　　要：针对融合多张部分聚焦图像的问题,从离焦图像质量评价角度出发,提出了一种基于视觉结构相似度的图像质量评价标准,并以此标准将离焦图像融合转化为一个优化问题.通过特殊设计进化算法,从源图像出发,优化出一个质量评价最好的图像作为融合结果.结果表明,通过该方法能够得到较好的融合结果,并在继承性和可靠性方面优于流行的多尺度方法和特征组合方法.When an image is captured by optical sensor,some parts of the scene will be blur due to lost-focus.Image fusion is a method for integrating multi-focus images into a composite image which is more suitable for work.In order to fuse multi-focus images,a new blind image quality assessment metric,Vision Structural Similarity（VSS）,was first proposed in this paper.Based on this metric, multi-focus image fusion problem could be described as an optimization process. Through optimizing source images by evolutionary computation,the image with best VSS value would be chosen as fusion result.The experimental results of the proposed method show high performance in both consistency and reliability comparing with other fusion schemes.

关 键 词：离焦图像融合 图像质量评估 视觉结构相似度 进化计算 

分 类 号：TP391[自动化与计算机技术—计算机应用技术] TN911.37[自动化与计算机技术—计算机科学与技术]