基于Retinex的弱可见光和红外图像融合算法

目的 针对目前弱可见光与红外图像融合后的图像仍存在细节大量丢失、目标模糊不清的问题，提出一种基于Retinex对弱可见光图像进行增强预处理后，再基于NSST和SWT变换进行图像融合的算法。方法 首先用SSR对弱可见光图像进行增强处理，增强后的可见光和红外图像进行NSST分解得到第1次的高低频系数，高频系数采用基于局部能量特征的方法进行融合；低频系数经过SWT分解得到第2次高低频系数，第2次的高频系数采用同样的方法融合，低频系数采用线性加权方法融合，然后将第2次高低频的融合结果经过SWT逆变换得到新的低频系数。最后把第1次高频系数融合结果和新的低频系数进行NSST逆变换得到融合图像。结果 通过仿真实验，将文中算法与NSST，NSCT以及文献5]算法进行对比，结果表明主观视觉上融合图像细节更加清晰，客观评价上，平均梯度、空间频率（SF）、标准差、信息熵、边缘信息保留量等指标分别提高了35.63%，26.73%，16.89%，7.2%，4.6%。结论 文中算法对图像融合有较好的改善作用，融合图像的可视性和图像质量都得到显著提高。

Fusion Algorithm of Low Visible Light and Infrared Image Based on Retinex

The work aims to propose an algorithm to enhance the weak visible light image based on Retinex and then to fuse image based on NSST and SWT for the problem that the images after the fusion of weak visible light and infrared images still have a large amount of details lost and blurred target. Firstly, the SSR was used to enhance the low visible light image, and then the enhanced visible and infrared images were decomposed by NSST to obtain the first high and low frequency coefficients. The high frequency coefficients were fused by a method based on local energy characteristics. The low frequency coefficients were decomposed by SWT to obtain the second high and low frequency coefficients. The high frequency coefficients of the second time were fused by the same method, and the low frequency coefficients were fused by the linear weighting method. Then, the low frequency coefficients of the second time were obtained through the inverse transformation of SWT. Finally, the fusion results of the high frequency coefficients of the first time and the new low frequency coefficients were transformed by NSST inverse transformation to obtain the fusion image. Through simulation experiments, the proposed algorithm was compared with NSST, NSCT, and the algorithm in literature 5. The results showed that the fusion image details were clearer in subjective vision, and the average gradient, standard deviation, information entropy and information retention indexes increased by 35.63%, 26.73%, 16.89%, 7.2% and 4.6% respectively in the objective. The proposed algorithm has a better effect on image fusion, and the visibility and quality of the fusion image are improved.