针对传统Retinex 算法处理红外图像存在光晕伪影和细节增强不足的缺点，本文提出一种消除光晕和细节增强的Multi-scale Retinex（MSR）红外图像增强算法。首先，以局部方差和局部复杂度构造引导滤波的自适应平滑增益，然后，采用改进的引导滤波核函数估计照度分量，在对数域对多尺度Retinex 数学模型求解，获取消除光晕和细节保持的多尺度反射分量。最后，为进一步增强细节和提升亮度，对反射分量依灰度等级进行自适应增强，并通过偏移调整和Gamma 校正改善图像亮度，得到最终增强图像。实验结果表明，本文算法相对其它的Retinex 增强算法，可有效地消除光晕现象，突出细节，可获得视觉效果良好的增强结果。

Aiming at the drawbacks of traditional Retinex for infrared image which suffer from the halo artifacts and the insufficient in detail enhancement, in this article, a multi-scale Retinex algorithm for infrared image which could remove halo and enhance details was proposed. Firstly, an adaptive smoothing gain was constructed by the local variance and the local complexity. Then the kernel function of improved guided filter was used to estimate the illumination component, the multi-scale Retinex mathematical model was solved in logarithmic domain, the multi-scale reflection components with Halo-free and detailpreserving properties were obtained. Finally, in order to enhance the detail and boost the brightness, reflection component was adaptively enhanced based on gray level, meanwhile, image brightness was improved via offset adjustment and Gamma correction, the final enhanced image was gained. The experimental results demonstrate that, compared with other Retinex enhancement algorithms, our method can efficiently reduce halo artifacts, highlight the detail, achieve a ‘visually pleasing’ enhancement result.