马尔科夫随机场模型下的Retinex夜间彩色图像增强

由于Retinex算法在处理夜间彩色图像时容易出现光晕、颜色失真、细节丢失与噪声干扰等问题，本文基于马尔科夫随机场（MRF）提出了一种针对单幅图像的Retinex图像增强算法。该算法在HSV颜色空间下采用线性引导滤波估计图像照度分量；在MRF模型下求解仅包含物体本身特性的反射分量，并通过颜色恢复函数与增益补偿方法进行颜色恢复与校正，最终实现了夜间彩色图像的增强。实验结果表明，利用本文算法处理后图像的均值（整体亮度）可以提高2倍以上，标准差、熵、峰值信噪比（PSNR）等参数均有5%以上的提升。与其它基于Retinex原理的算法相比，本文提出的算法增强效果显著，具有消除“光晕伪影”现象、抑制噪声、颜色保真和有效地凸显边缘细节信息等能力。

A Retinex algorithm for night color image enhancement by MRF

As Retinex algorithm usually has the problems of halo artifacts,color distortion, high noises and poor details in a low-illumination night color image, this paper proposes a novel Retinex enhancement algorithm based on Markov Random Fields (MRF) to enhance the visibility of single image. This algorithm uses the linear guided filter to estimate image illumination component in HSV(Hue Saturation Value) color space. The reflection image can be obtained through MRF model, and the night color image can be enhanced after color restoration and brightness correction. Compared with the original images, the experimental results demonstrate that the mean value (luminance) of image restored by proposed method has increased more than 2 times, and the evaluation indexes such as standard deviation, entropy, the Peak Signal and Noise Ratio(PSNR) and so on have increased more than 5%. Compared with other Retinex algorithms, the effect of enhancement of this algorithm is more remarkable in halo effect elimination, noise suppression and detail preservation.