一种新型的雨雪视频图像复原方法

针对雨、雪等恶劣天气极易影响视频图像质量的问题,提出了一种新型雨雪视频图像复原方法。该方法在深入研究视频帧序列时域特性的基础上,将R、G、B 3维彩色空间属性数据进行投影降维,通过研究低维数据的散布形状来揭示高维数据的聚类特征,进而依据统计特性进行雨雪去除,实现视频图像复原的目的。该方法依据雨雪区和背景区的数据特性进行分类,能提供更准确的类别中心,尤为关键的是,该方法所需视频序列帧数少,不需要进行数据的彩色空间转换,简化了运算。6个经典视频场景的实验结果表明,复原后的图像清晰度高,具有较强的鲁棒性。

A Novel Restoration Algorithm of Rain and Snow from Videos

In order to reduce the effect of the dynamic weather on the performance of outdoor vision system in videos such as rain and snow, a novel algorithm of image restoration for rain and snow was proposed. Based on the properties of RGB, the proposed algorithm worked only on main characteristics by projection high dimensional data to low dimensional space. Rain and snow were identified accurately. Also, fewer frames of videos and non transformation from one color space to another was simple to use. The effectiveness of our algorithms was demonstrated by six classical scenes. Experimental results showed that it could effectively discriminate the rain pixels and non-rain pixels.