【摘要】 为解决弱光环境下的红外与可见光图像的融合结果存在对比度低、细节纹理不足、融合耗时长的问题，提出了一种改进多尺度结构化融合的方法。在图像融合前，采用动态范围压缩算法对弱光下的可见光图像进行增强，再通过多尺度结构化分解将增强后的可见光图像和红外图像分解成低频和高频信息；在融合过程中，提出一种基于均方根误差系数的方法对低频信息进行融合，提出一种基于信息熵自适应调整权重的策略对初步融合的高频信息进行二次优化融合，再通过多尺度结构化分解的逆变换重构出融合图像；最后，提出一种基于灰度分类的区域像素增强算法来提高融合后图像的对比度。将提出方法与9种常用的融合方法进行了定性和定量的分析比较，在TNO和CVC-14数据集上的实验结果表明，该方法在平均梯度、交叉熵、边缘强度、标准差和空间频率指标上取得了更好的客观评价结果，整体视觉效果也要优于对比方法。本方法的融合结果具有丰富的细节纹理、较高的清晰度和对比度，且融合耗时短。更多还原

【Abstract】 Under low-light conditions, the fusion of infrared and visible images often results in images with poor contrast, lacking in detail, and requiring a lengthy processing time. To address these issues, this paper introduces an enhanced multi-scale structural fusion approach. Initially, it improves the low-light visible image using a dynamic range compression enhancement algorithm. Subsequently, through a multi-scale structural image decomposition method, it separates the enhanced visible and infrared images into their low-frequency and high-frequency components. For image fusion, the low-frequency components of both image types are merged using a technique based on the root mean square error coefficient. The high-frequency components are initially fused in a straightforward manner, followed by an optimized fusion using a self-adaptive weight adjustment based on image information entropy. Afterward, by reversing the multiscale structural decomposition, the fused low and high-frequency components are combined to form a complete image. To further enhance the image contrast, a regional pixel enhancement algorithm based on gray level classification is introduced. The effectiveness of this method is compared with nine conventional infrared and visible image fusion techniques, both qualitatively and quantitatively, using TNO and CVC-14 datasets. The proposed method demonstrates superior performance in terms of average gradient, cross entropy, edge intensity, standard deviation, and spatial frequency, along with an improved overall visual quality. This confirms that the images produced by the proposed method exhibit enhanced detail, clarity, contrast, and are processed more quickly.更多还原