融合多帧信息的Retinex监控视频增强算法

Integration of multi-frame information Retinex surveillance video enhancement algorithms

针对传统Retinex监控视频增强算法照度分量提取不够准确的情况,利用监控视频背景的时间相关性,融合多帧背景进行照度估计,提出了一种新型Retinex监控视频增强算法:对视频单帧图像进行大、中、小尺度的高斯低通滤波,得到3个尺度的环绕图像,并对其取极小值以提取该帧的背景照度图像,通过融合当前帧和其邻帧的多帧背景照度图像,获取当前帧上准确的背景照度图像,再应用Retinex色彩恒常性理论,去除照度干扰以获得反射光分量,实现当前帧的增强。实验结果表明:该算法可以从夜间视频阴影中恢复出景物,得到亮度、色彩、细节较平衡的视频。

The luminance component extraction of traditional Retinex surveillance video enhancement algorithm is often not accurate enough.In view of the similarity of background and illumination between frames of surveil-lance videos,with rational fusing of the information between frames,an enhancement algorithm for surveillance videos based on multi-scale Retinex is proposed.This algorithm uses low pass Gauss filters with large,medium and small three different scales for signal frame video to get three dimensions around the image,and extracts the background illumination image to the minimum value.By the fusion of the current frame and its adjacent frame background illumination image,more accurate background illumination image on the current frame can be ob-tained,and then Retinex theory of color constancy can be used to get rid of the illumination interference and get the reflection components.The image enhancement of current frame can be realized.The experiment results show that the algorithm can recover scenery from video shadows at night and get the balance video with bright-ness,color and details.