摘要
由于缺乏平移不变性,Contourlet变换用于图像融合时,如果源图像未完全配准,融合后图像会出现伪吉布斯现象,而常用的解决方法——非下采样Contourlet变换运行速度较慢.为解决上述问题并提高图像融合质量,提出一种基于人眼视觉系统特性和循环平移Contourlet变换的图像融合新方法.方法中,利用循环平移克服融合过程中奇异点产生的伪吉布斯现象,并对Contourlet变换后的融合规则进行改进,在低频和高频的处理过程中,充分考虑人眼视觉特性,提出低频运用局部纹理特征选择融合系数,高频运用视觉显著度选择融合系数.通过对不同融合方法进行仿真和真实多聚焦图像融合实验,并从主观和客观两方面进行评价,实验结果表明本文方法均优于传统的小波变换、移不变小波变换、Contourlet变换、非下采样Contourlet变换等融合方法,运行效率比非下采样Contourlet变换更高.
Due to lack of shift invariance, when the contourlet transform is used for image fusion, the fused image tends to have the Gibbs-like phenomenon if the source images are not completely registered. The Non-subsample contourlet transform （NSCT） has already been proposed to overcome this shortcoming, however, the computing efficiency of the method is slow. In order to solve the above problems and improve the fusion quality, this paper presents a novel multi-focus image fusion method based on Human Visual System （ HVS ） and cycle spinning contourlet transform. The cycle spinning technique has been employed to overcome the Gibbs-like phenomenon, which comes from the singular points in fusion process. Moreover, an improved fusion rule has been put forward after the contourlet decomposition. In our method, the low frequency and high frequency sub-bands are handled differently according to HVS characteristics, i. e. , the coefficients of low frequency are selected by using a local textural features based rule, whereas the coefficients of high frequency are fused via a visual saliency based mechanism. Experimental results on both synthetic images and real multi-focus images indicate that the proposed method is superior to series of fusion methods, including conventional discrete wavelet transform （ DWT）, shift-invariant DWT （ SIDWT）, contourlet transform and NSCT, in terms of both subjective and objective evaluations. In addition, the efficiency of the proposed method is higher than that of NSCT.
出处
《小型微型计算机系统》
CSCD
北大核心
2016年第6期1348-1354,共7页
Journal of Chinese Computer Systems
基金
国家自然科学基金项目(61262034
61462031)资助
教育部科学技术研究重点项目(211087)资助
江西省自然科学基金项目(20151BAB207033)资助
江西省高校科技落地计划项目(KJLD14031
GJJ150438)资助
