基于二维经验模态分解和高斯模糊逻辑的红外与可见光图像融合

Fusion of infrared and visible images based on BEMD and GFL

针对传统多尺度图像融合方法容易损失可见光图像细节、弱化红外目标信息和降低图像对比度的问题,基于二维经验模态分解(BEMD)和高斯模糊逻辑(GFL)的特性,提出了一种红外与可见光图像融合的算法。首先,使用BEMD对源图像进行分解,得到图像的本征模(高频成分)和趋势项(低频成分);其次,用GFL对趋势项进行恰当的融合,使用基于邻域特征的区域对比度法融合图像的本征模;最后,通过BEMD逆变换得到融合图像。实验结果表明,与传统的多尺度融合方法相比,在主观上视觉上,本文融合算法能够更有效地保留源可见光图像中的细节信息,并突出红外图像中的目标信息,提高融合图像的质量;在客观评价指标上,本文融合算法的结果在信息熵(IE)、标准差(SD)、平均梯度(AG)、互信息(MI)和空间频率(SF)5个客观指标上明显优于传统的多尺度融合方法。

The traditional image fusion methods based on multi-scale transform can easily loss the details of visible image, weaken the infrared target information and reduce the image contrast. To solve the problem,a fusion method of infrared and visible images is proposed based on bidimensional empirical mode decomposition （BEMD）and Gaussian fuzzy logic （GFL）. Firstly, the source images are decomposed to obtain intrinsic modes （high-frequency component） and trend terms （low-frequency component）by BEMD. Secondly,trend terms are combined properly by using the GFL based fusion rule,and intrinsic modes are merged by using regional contrast method based on neighborhood characteristics. Finally, the fusion image is obtained by performing the inverse BEMD transform on the combined coefficients. The experimental results prove that the proposed method outperforms the traditional multi-scale methods for infrared-visible images fusion. Subjectively, the proposed fusion method can improve the quality of fusion image by preserving the details of source visible image and highlighting the targets of source infrared im- age. Objectively,the results of the proposed fusion method are obviously better than those of the traditional multi-scale methods in the objective evaluation values,such as information entropy （IE）, standard deviation （SD）,average gradient （AG）,mutual information （MI） and spatial frequency （SF）.