Infrared and visible image fusion method based on compound decomposition and intuitionistic fuzzy set

基于复合分解与直觉模糊集的红外与可见光图像融合方法

¹ School of Mathematics and Statistics, Shaanxi Xueqian Normal University, Xi'an 710100, China
² School of Computer Science, Northwestern Polytechnical University, Xi’an 710072, China

Received: 8 December 2020

Abstract

To overcome the shortcomings of traditional image fusion algorithms based on multiscale transform, an infrared and visible image fusion method based on compound decomposition and intuitionistic fuzzy set is proposed. Firstly, the non-subsampled contour transform is used to decompose the source image into low-frequency coefficients and high-frequency coefficients. Then the potential low-rank representation model is used to decompose low-frequency coefficients into basic sub-bands and salient sub-bands, in which the visual saliency map is taken as weighted coefficient. The weighted summation of low-frequency basic sub-bands is used as the fusion rule. The maximum absolute value of low-frequency salient sub-bands is also used as the fusion rule. The two fusion rules are superimposed to obtain low-frequency fusion coefficients. The intuitionistic fuzzy entropy is used as the fusion rule to measure the texture information and edge information of high-frequency coefficients. Finally, the infrared visible fusion image is obtained with the non-subsampled contour inverse transform. The comparison results on the objective and subjective evaluation of several sets of fusion images show that our image fusion method can effectively keep edge information and rich information on source images, thus producing better visual quality and objective evaluation than other image fusion methods.

摘要

针对传统的基于多尺度变换的红外与可见光图像融合算法的不足，提出一种基于复合分解与直觉模糊集的红外与可见光图像融合方法。采用NSCT将源图像分解为低频子带和高频子带，进一步采用潜在低秩表示模型将低频子带分解为低频基础子带和低频显著子带；针对低频基础子带、低频显著子带和高频子带的特征，采用不同的融合规则，其中，低频基础子带以视觉显著度为权重系数采用加权求和作为融合规则，低频显著子带以绝对值最大为融合规则，高频子带以直觉模糊熵最大选择为融合规则；通过NSCT逆变换得到红外与可见光融合图像。通过对比多组融合图像主、客观评价结果表明，该方法能有效保留边缘信息，保留较多的源图像信息，在视觉质量和客观评价方法优于其他图像融合方法。