基于改进snake模型的图像中雨雪去除算法研究*

摘要: 针对雨雪在图像处理中造成的不利影响，提出一种基于改进snake模型的雨雪去除新方法。传统snake模型的初始轮廓点为手工确定，且只适用于目标边缘清晰的情况，而雨雪的轮廓并不明显，因此该算法利用模糊连接度自动确定有序初始轮廓点；由于高速下落的雨雪在图像中形成模糊边缘，传统的snake模型不能准确地收敛到边界点，该算法利用模糊相似度函数来构造snake模型的外部能量函数，从而准确定位雨雪边界，而后采用三次B样条得到连续光滑的雨雪轮廓；并且为了消除运动因素的干扰，应用了HSI颜色空间的H元素。通过对大量雨雪图像进行实验，结果表明，该算法可以有效地识别出不同降速的雨滴或雪片，并能较好的消除移动物体的影响。

Research on removal algorithm of rain and snow from images based on improved snake model

As for the adverse effects of rain and snow on image processing, this paper proposed a new removal algorithm for rain and snow based on improved snake model. Normally, conventional snake model determined the initial contour points by hand, which only applied to clear target edge. Aiming at the non-obvious outline of rain and snow, this algorithm could automatically obtain the initial sequential contour points using fuzzy connectedness. Moreover, using conventional snake model, blurred edge induced by high-speed drop of rain or snow cause that the initial contour points could not accurately converge to the boundary points. Therefore, the algorithm utilized fuzzy similarity function to construct the external energy function of snake model in order to locate the boundary of rain and snow precisely, and then smoothed and fitted the profile of rain and snow through cubic B-spline. In addition, applied the H component of the HSI color space to reduce the impact of moving objects on rain and snow removal. A lot of experiments were processed. And the results show that the proposed algorithm is more suitable to identify rain or snow with different velocities, and better to eliminate the impact of moving objects.