【摘要】 为了克服Contourlet融合在远离支撑区间上出现的混叠成分,抑制融合图像在奇异处产生伪吉布斯现象,提出了改进拉普拉斯能量和的尖锐频率局部化Contourlet(SFLCT)域多聚焦图像融合方法。采用SFLCT而不是原始的Cont-ourlet对多聚焦图像进行分解,并将多聚焦图像空域融合方法中评价图像清晰度的指标引入到SFLCT变换域,用拉普拉斯能量来选择变换域系数。然后,逆SFLCT重构得到融合结果。最后,采用循环平移来提高SFLCT的平移不变性,有效抑制融合图像在奇异处产生伪吉布斯现象。实验结果表明:对于多聚焦图像,所提方法比循环平移小波变换法的互信息提高了5.87%,QAB/F提高了2.70%,比循环平移Contourlet方法的互信息提高了1.77%,QAB/F提高了1.29%;视觉效果优于典型的空域分块拉普拉斯能量方法和平移不变小波变换方法。更多还原

【Abstract】 In order to suppress the pseudo-Gibbs phenomena around singularities of fused images and to reduce significant amounts of aliasing components located far away from desired supports when the original Contourlet is employed in the image fusion,a multifocus image fusion method in Sharp Frequency Localized Contourlet Transform(SFLCT) domain based on a sum-modified-Laplacian is proposed.The SFLCT,instead of the original Contourlet,is utilized as the multiscale transform to decompose the original multifocus images into subbands.Then,typical measurements for the multifocus image fusion in a spatial domain are introduced to the Contourlet domain and Sum-modified-Laplacian(SML),and the criterion to distinguish SFLCT coefficients from the clear parts or from blurry parts of images are employed in SFCLT subbands to select the SFLCT transform coefficients.Finally,the inverse SFLCT is used to reconstruct fused images.Moreover,a cycle spinning method is applied to compensate for the lack of translation invariance property and to suppress the pseudo-Gibbs phenomena of fused images.Using the proposed fusion method,experimental results demonstrate that the mutual information has improved by 5.87% and transferred edge information QAB/F has improved by 2.70% as compared with those of the cycle spinning wavelet method,and has improved by 1.77% and 1.29% as compared with those of the cycle spinning Contourlet method.Meanwhile,the proposed fusion method has advantages of good visual effect over the block-based spatial SML method and shift-invariant wavelet method.更多还原