Abstract: A novel fast algorithm based on Retinex for video enhancement is presented，to direct at the long run time of existent algorithm.This method uses the Retinex algorithm twice to divide each frame into global illumination，local illumination and the reflection.The local illumination includes local illumination variety caused by subject’s motion.This method can preserve the motion without complex motion estimation.Results on a series of night video show the efficiency of the approach.

Key words: video enhavement, Retinex algorithm, bilateral filter, temporal filter

摘要： 针对现有视频增强算法时间复杂度较高的局限性，提出一种新的快速Retinex视频增强算法。该算法通过两次Retinex算法，将视频每一帧分解成照度分量和反射分量，再将照度分量分解成全局照度和局部照度。其中全局照度包含场景光照信息，局部照度包含由物体运动产生的光照变化。该算法可通过局部照度保留物体的运动信息，避免“运动伪影”，计算复杂度较低。实验表明，该算法对夜间高动态范围交通视频，具有良好的增强效果，且运行速度较快。

关键词: 视频增强, Retinex算法, 双边滤波, 时域滤波