计算机应用 ›› 2011, Vol. 31 ›› Issue (11): 3022-3026.DOI: 10.3724/SP.J.1087.2011.03022

• 图形图像技术 • 上一篇    下一篇

基于视觉注意机制与区域结构相似度的图像融合质量评价

任仙怡1,2,刘秀坚1,3,胡涛1,2,张基宏1,2,3   

  1. 1. 深圳市可视媒体处理与传输重点实验室,广东 深圳 518029
    2. 深圳信息职业技术学院 信息技术研究所,广东 深圳 518029
    3. 深圳大学 计算机与软件学院,广东 深圳 518060
  • 收稿日期:2011-05-30 修回日期:2011-07-08 发布日期:2011-11-16 出版日期:2011-11-01
  • 通讯作者: 任仙怡
  • 作者简介:任仙怡(1973-),女,河南洛阳人,副教授,博士,主要研究方向:模式识别、图像处理、机器视觉、视频监控;刘秀坚(1983-),男,广东梅州人,硕士研究生,主要研究方向:数字图像融合;胡涛(1979-),男,湖北黄冈人,讲师,博士,主要研究方向:图像处理、机器视觉;张基宏(1964-),男,江苏海安人,教授,博士生导师,主要研究方向:数据压缩、数字信号处理。
  • 基金资助:
    国家自然科学基金资助项目;广东省自然科学基金资助项目

Objective quality evaluation of image fusion based on visual attention mechanism and regional structural similarity

REN Xian-yi1,2,LIU Xiu-jian2,3,HU Tao1,2,ZHANG Ji-hong1,2,3   

  1. 1. Institute of Information Technology, Shenzhen Institute of Information Technology, Shenzhen Guangdong 518029, China
    2. Shenzhen Key Laboratory of Visual Media Processing and Transmission, Shenzhen Guangdong 518029, China
    3. College of Computer Science and Software Engineering, Shenzhen University, Shenzhen Guangdong 518060, China
  • Received:2011-05-30 Revised:2011-07-08 Online:2011-11-16 Published:2011-11-01
  • Contact: REN Xian-yi

摘要: 针对目前图像融合质量客观评价与主观评价结果一致性不高的问题,结合人类视觉注意机制和区域计算方法对EFQI指标进行改进,提出了一种基于视觉注意机制(VAM)与区域结构相似度的图像融合质量评价指标。该指标充分考虑了人类视觉感兴趣区域和人眼对区域信息敏感的特性,使用方差显著图与视觉显著图相结合的加权方法对图像中的显著区域赋予更大的权值,并采用更符合人类视觉特性的区域计算方法求取融合结果图像和源图像在各区域上的结构相似度以评价融合质量。用该指标评价10种不同融合算法的融合结果图像并与主观评价结果进行相关性分析,分析结果表明该方法相比传统的评价指标能更有效地反映融合图像质量,与主观评价结果一致性更高。

关键词: 图像融合, 质量评价, 主观评价, 客观评价, 视觉注意机制, 区域结构相似度

Abstract: To handle the problem of low consistency between the objective and subjective evaluations of image fusion, considering the features of Human Visual System (HVS), a new metric to evaluate the quality of the fusion image based on the Visual Attention Mechanism (VAM) and the regional structural similarity was proposed. This quality metric utilized the global salience got by VAM and the local salient information to estimate how well the salient information contained within the sources was presented by the composite image. Since human eyes are more sensitive to region, by giving higher weight to those regions with high saliency value in the source images, the new metric evaluated the quality of the fused image by computing the weighted regional structural similarity of the fused image and source images in all regions. The correlation analysis between objective measure and subjective evaluation was performed and the results demonstrate that the new metric is more consistent with human subjective evaluation, compared with the traditional objective measurements and the widely used EFQI.

Key words: image fusion, quality evaluation, subjective evaluation, objective evaluation, Visual Attention Mechanism (VAM), regional structural similarity