图像融合数据集

图像融合数据集主要面向低照度隧道口视觉融合技术研究，解决隧道出入口光照强度不足导致成像质量不佳的问题，基于低照度隧道口红外与可见光图像融合方法。首先利用双边滤波与光照分量，对隧道口低照度红外和可见光源图像进行自适应图像增强；其次通过非下采样轮廓波进行多尺度、多方向分解以弥补预处理后的图像信息损失；在低频系数上，采用基于卷积稀疏表示与局部能量特征相结合的方法进行融合；在高频系数上，根据底层视觉特征构建新活性度量方法与光谱边缘处理；最后，将得到的低频和高频融合层进行重构得到最终的融合图像。实验结果表明，所提出的融合算法与BF、SE、NSCT-BF、SF-Energy-Q、SR-C&L五种算法相比，主观评价上视觉效果更好，辨识度高，整幅图像场景得以凸显，互信息量、信息熵、标准差均为最高，分别为7.5962、7.7642、82.1941，运算时间至多减少0.0232 s。该方法在降低噪声、均衡光照、恢复细节方面有参考意义。

This image fusion dataset is mainly developed for research on visual fusion technology at low-light tunnel entrances and exits, aiming to solve the problem of poor imaging quality caused by insufficient light intensity at tunnel entrances and exits, and is based on infrared and visible light image fusion methods for low-light tunnel entrances and exits. First, adaptive image enhancement is performed on low-light infrared and visible light images of tunnel entrances and exits using bilateral filtering and illumination components. Second, multi-scale and multi-directional decomposition is carried out via Non-Subsampled Contourlet Transform (NSCT) to compensate for image information loss after preprocessing. For low-frequency coefficients, a fusion method combining convolutional sparse representation and local energy features is adopted. For high-frequency coefficients, a novel activity measurement method and spectral edge processing are constructed based on low-level visual features. Finally, the obtained low-frequency and high-frequency fusion layers are reconstructed to obtain the final fused image. Experimental results show that compared with five baseline algorithms including BF, SE, NSCT-BF, SF-Energy-Q and SR-C&L, the proposed fusion algorithm achieves better visual effect in subjective evaluation, with high distinctiveness and prominent overall image scene. Its Mutual Information (MI), Entropy and Standard Deviation are the highest, reaching 7.5962, 7.7642 and 82.1941 respectively, and the computation time is reduced by up to 0.0232 s. This method has reference significance for noise reduction, illumination equalization and detail restoration.

搜集汇总

背景与挑战

背景概述

该数据集专注于隧道口低照度环境下的红外与可见光图像融合技术研究，提供了一套包含572个文件、总计225.5MB的数据资源，用于支持图像增强和融合算法的开发与验证。